科資中心編號: RN9101-0236 
題名: 多重機器人系統之研製 
Design and Implementation of Multi-Agent Robotic System 
研究者: 杜國洋 
機構名稱: 華夏工商專校電子工程科 
經費來源: 行政院國家科學委員會 
補助編號: NSC89-2218-E146-003 
研究日期: 2000/08 - 2001/07 
出版日期: 2001/07 
頁冊數: 4頁 
原件提供: 行政院國家科學委員會科學技術資料中心, NSC89-2218-E146-003 
科資分類號: IF0303 
關鍵詞: 人工智慧 ;多重機器人系統 ;機器足球比賽 ;工業標準架構 
Artificial Intelligence (Ai) ;Multi-Agent Robotic System ;Robocup ;Industry Standard Architecture (Isa) 
英文摘要: RoboCup is designed to evaluate various theories, algorithms, and architectures for new challenges of Artificial Intelligent (AI) research [1]. New challenges will import new functions into a soccer robot. Thus, a soccer robot, designed as an embedded system with modularized control circuits is proposed to conveniently add new functions for further challenges. At present, the designed soccer robot, based on circuit function, has five modules: processor module, radio communication module, input/output module, supersonic module and extensible module. Five modules are together connected via Industry Architecture Standard bus (ISA bus) in extensible module. In addition, the motion planning of a soccer robot for kicking the ball is formularized to shoot goal. Experiment is also included. 



科資中心編號: RB9102-0792 
題名: 運用智慧型行動代理人技術於商業流程 
Intelligent Mobil Agent for Business Process Management 
研究者: 郭忠義 
機構名稱: 輔仁大學資訊工程系 
經費來源: 行政院國家科學委員會 
補助金額: 新台幣 284 仟元 
補助編號: NSC89-2218-E030-006 
研究日期: 2000/11 - 2001/07 
出版日期: 2001/10 
頁冊數: 5頁 
原件提供: 行政院國家科學委員會科學技術資料中心, NSC89-2218-E030-006 
科資分類號: IA0407;IA0301 
關鍵詞: 行動代理人 ;商務流程 ;文件驅動法 ;網際網路 ;工作流 ;人工智慧 
Mobile Agent ;Business Process ;Document Driven Approach ;Internet ;Workflow ;Artificial Intelligence (Ai) 
中文摘要: 工作群組中為達成某一共同目的而需要多人協力以循序或平行工作的形式共同完成任務,此系統提供控制、監督以及管理整個內部流程的能力,稱為工作流程管理系統。隨著網際網路蓬勃發展,越來越多的工作流程管理系統應用於處理企業內部的商業流程,尤其以網路連結的跨部門甚至跨地區的企業應用系統,更需要一套系統控制、監督與管理。近年來,代理人的獨立自主性與彈性互動,使得代理人技術於商業流程管理領域的研究應用越來越受重視,但大部分研究卻都忽略代理人的行動性與智慧力。本研究計畫基於人工智慧與行動式代理人技術改良多重代理架構的進階工作決策環境方法論(APEPT),成為行動式代理人決策方法論(MADT)。行動式技術的移動性使代理人的溝通更具彈性,包括改良ADEPT外部架構以減低網路負荷;人工智慧技術則增強代理人生命週期與環境的認知能力,包括改良代理人內部架構。MADT 是文件驅動的代理人系統,將文件封裝為文件代理人達成(1)可追蹤能力:系統管理者容易對文件流程監控;(2)文件之完整生命週期:利用行動代理人的生命週期管理文件的生命週期,並且達成並行處理;(3)動態排程:文件代理人可視環境變動,動態改變文件流程,提供不同服務的排程。本研究計畫的主要目的在運用人工智慧與行動式代理人技術發展一個文件驅動的代理人系統,應用於商業流程管理系統達成:(1)提供執行商業流程的環境並根據限制條件,讓商業流程準時完成;(2)對商業流程的運作提供動態排程;(3)監視商業流程的運作;(4)管理文件的生命週期。 
英文摘要: A workflow is a composite activity consisting of tasks involving control, monitoring and management the internal process. Due to the development of Internet and the desires of almost each department to be interconnected and needs to make data accessible at any time and any place, more and more workflow management systems are applied to business process management. The software agent has been widely adopted in the applications. However, most researchers do not propose mobility to developing an agent system. In addition, due to the different properties from the communication environment, a artificial intelligence technology for an agent system is in need. In this project, a mobile, intelligent and document-driven agent system is proposed to modeling business process management system. Our approach includes these major parts: (1) to monitor the business process; (2) to trace the document process; (3) to dynamic schedule document agent itinerary; (4) to manage document life cycle. 


[科資中心編號]RB8703-0080
[題    名]
以基因演算法製作投資決策支援系統之知識庫架構研究
A Genetic-Based Technique for the Automated Acquisition of the
Knowledge Bases for Trading Systems
[研 究 者 ]
陳安斌
Chen, An-Pin
[機構名稱  ]交通大學資訊管理研究所(NCTUIFMI)
[經費來源  ]行政院國家科學委員會(NSC)
[補助金額  ]221000
[補助編號  ]NSC86-2416-H009-005
[研究開始日期]1996/08
[研究結束日期]1997/07
[頁 冊 數 ]84頁
[語    文]英文
[建檔日期  ]1999/02
[出版情形  ]完成日期:民國87年5月
[原件提供  ]行政院國家科學委員會科學技術資料中心
NSC86-2416-H009-005
[科資分類號 ]IB0002;IA0407
[關 鍵 詞 ]
遺傳演算法;交易策略;人工智慧;決策支援系統;系統模擬;知識庫

Genetic algorithm;Trading strategy;Artificial intelligence;Decision
support system (DSS);System simulation;Knowledge base

[摘    要]
本報告的目標希望建立一基因演算法為基礎的交易知識庫。首先建立交易規則的基因表示方法必須被建立,繼而以基因演算法產生交易規則,然後,這些規則將以事先定義妥的適化函式評估績效,最後,以基因演算法最佳化的自動交易系統將藉此建立。根據本報告的結果顯示,以基因演算法所建立的自動交易系統特別適用於找出關鍵規則模式,包含這些規則的參數。

The major objective of this paper is to propose a model to
construct knowledge bases of trading systems by genetic algorithms.
First of all, both of the knowledge representation method and
grammars needed for trading rules will be defined. The second step is
that the trading rules will be generated by genetic algorithms. And
in the third stage, these rules will be evaluated by the pre-defined
fitness function and the qualified rules will be extracted. Finally,
a complete trading systems along with the qualified rules would be
demonstrated.
According to the results of this paper, it can be concluded
that using genetic algorithms to construct the knowledge base of the
trading system is particularly efficient in finding out the key rule
models, including the necessary parameters of the rule models. And
the simulation result shows that the performance of these rules are
quite excellent in both profit margin and risk control.

[科資中心編號]RA8604-2064
[題    名]
貝氏網路系統之發展及在不確定性推理之應用
Design of Bayesian Network for Uncertain Reasoning.
[研 究 者 ]
李宗南
LEE,CHUNG-NAN
[機構名稱  ]中山大學資訊工程研究所(NCSUCIEG)
[經費來源  ]行政院國家科學委員會(NSC)
[補助金額  ]136000
[補助編號  ]NSC84-2213-E110-019
[研究開始日期]1995/02
[研究結束日期]1995/07
[頁 冊 數 ]117頁
[語    文]中文
[建檔日期  ]1997/11
[出版情形  ]完成日期:民國84年8月
[原件提供  ]行政院國家科學委員會科學技術資料中心
NSC84-2213-E110-019
[科資分類號 ]IA0407;IA0202
[關 鍵 詞 ]
人工智慧;Bayesian網路;不確性推理;圖形使用者介面;物件導向

Artificial intelligence;Bayesian network;Uncertainty reasoning;
Graphic user interface;Object orientation

[摘    要]
在處理不確定性推理(Uncertainty reasoning)的問題時,貝氏網路
(Bayesian network)已經被認為是一個很有效的工具,在近幾年來已被廣泛的應用在人工智慧(AI)中有關不確定性推理的領域,如醫學診斷、電腦視覺、天氣預報等等。為了讓研究不確定性推理學者能夠方便地使用貝氏網路的理論來解決問題,我們設計一個貝氏網路發展系統來做為處理不確定性推理的工具。這個系統是透過一個圖形使用者介面(GUI)來建立貝氏網路和輸入資料,並且可以隨時更新
資料並顯示推理的狀態。以物件導向(Object oriented)的觀念來分析和設計這個系統,包括GUI的設計、定義類別(Class)、找出各個類別彼此之間的關係,並且撰寫使用者手冊、文件(Documentation),測試軟體的正確性。最後我們用幾個例題來說明如何應用此系統,由例題的結果顯示本系統不但為使用者提供了一個友善的(Friendly)環
境來建立貝氏網路,而且也具有極強的推理能力。

Bayesian networks have been proved to be an effective tool for
tasks requiring automated reasoning under uncertainty and have been
applied to many research areas such as medical diagnosis, map
learning, language understanding, computer vision etc., in the last
few years. In order to let the Bayesian networks is accessible by
related researchers without struggling in understanding the
complicated theory and going through the tedious implementation, we
have developed and designed a set of tools for the establishment of
Bayesian networks. The networks is interactively created by adding
nodes and links through a graphical user interface. The belief fusion
and propagation is handled by updating the new evidence for certain
nodes, then the information of new state of network is computed and
displayed on the system.
The object oriented approach is used to develop the system. It
includes the graphical user interface, classes, objects, the
relationships among objects users' manual, documentation, testing,
and verification. A couple of examples, include medical diagnosis and
diagnosis of logic circuits, have been used to demonstrate the
powerful reasoning ability under new evidence and friendly
environment of the system.

[科資中心編號]RA8510-2135
[題    名]
一個以角度參數為基礎的證據推理模式之探討
Investigation of an Angular-Parameter-Based Evidential Reasoning
Model.
[研 究 者 ]
鄧洪聲
DON,HON-SON
[機構名稱  ]中興大學電機工程系(NCHSELE)
[經費來源  ]行政院國家科學委員會(NSC)
[補助金額  ]139000
[補助編號  ]NSC84-2213-E005-017
[研究開始日期]1995/02
[研究結束日期]1995/07
[出版日期  ]1995/07
[頁 冊 數 ]37頁
[語    文]英文
[建檔日期  ]1997/05
[出版情形  ]研究報告,37頁,民國84年7月
Report, 37p., Jul. 1995
[原件提供  ]行政院國家科學委員會科學技術資料中心
NSC84-2213-E005-017
[科資分類號 ]IA0407
[關 鍵 詞 ]
人工智慧;專家系統;證據推理;相信函數;相依
關係;證據結合

Artificial intelligence;Expert system;Evidential reasoning;Belief
function;Dependency relationship;Belief combination

[摘    要]
我們提出一種新的證據推理模式。在這個
模式中,每一個證據是以一個相信函數來表示,
這個相信函數可以是分離式的,也可以是連續
式的。我們考慮使用一個角座標的參數來表示
兩個證據之間的相依關係。我們在理論上分析
這種模式的穩定性及適用性,並且探討在這種
模式中各種證據推理的運算如何進行,這些運
算包括證據之結合、證據之傳播、且/或運算
、以及在一個規則中之傳播等。我們並且解決
了一個存在已久的問題,也就是當有許多證據
存在時,它們的結合次序會影響到最後的結果
這個問題。我們做的一些探討顯示這個模式在
解決上述問題上比現有的一些模式好很多。我
們以實驗驗證,以對這個模式的特性有所瞭解


We propose a new evidential reasoning model in which a piece
of evidence is represented by a belief function. The belief function
can be in discrete or continuous form. We consider to use an angular
parameter to represent the mutual dependency relationship between
two pieces of evidence. We have theoretically analyzed the stability
and feasibility of this model, and investigated how the various
operations of evidential reasoning are performed. These operations
include belief combination, belief propagation, AND/OR, and
propagation within a rule, etc. We have also solved a long existing
problem, i.e., the conflict resulting from the combination order
when multiple evidence exist. Our investigation shows that the
proposed model has great advantages over the conventional approaches
in solving above problem. Some experiments have been conducted to
demonstrate the properties of this model.

[科資中心編號]RA8510-2115
[題    名]
限制條件推論系統之研製
Study on the Constraint Reasoning System.
[研 究 者 ]
許清琦
HSU,CHING-CHI
[機構名稱  ]台灣大學資訊工程研究所(NTUGCSIG)
[經費來源  ]行政院國家科學委員會(NSC)
[補助金額  ]176000
[補助編號  ]NSC84-2213-E002-031
[研究開始日期]1995/02
[研究結束日期]1995/07
[出版日期  ]1995/07
[頁 冊 數 ]81頁
[語    文]英文
[建檔日期  ]1997/05
[出版情形  ]研究報告,81頁,民國84年7月
Report, 81p., Jul. 1995
[原件提供  ]行政院國家科學委員會科學技術資料中心
NSC84-2213-E002-031
[科資分類號 ]IA0407
[關 鍵 詞 ]
人工智慧;推理系統;限制滿足問題

Artificial intelligence;Reasoning system;Constraint satisfaction
problem

[摘    要]
限制滿足問題(CSP,Constraint satisfaction
problem)主要是依據變數間彼此的限制,在變數本
身所允許的值域中找到符合所有條件的解。它
可以廣泛地被運用在電腦視覺(Machine vision)、
排程(Scheduling)、時間推理(Temoral reasoning)以及
自然語言認知(Natural language understanding)等不同
範疇的領域中。在這論文中,我們將討論一套
良好的限制條件推理環境---CREDIT(Constraint
reasoning environment in Taiwan-university)如何來處理
大多數的限制滿足問題。
我們設計了一套問題描述語言來表達限制
滿足問題的模型。它主要的目的是作為使用者
和系統間良好的溝通介面。可以輕易的將使用
者所欲解之問題的重要條件如變數、變數值域
及彼此限制條件等描述完全。它可以表示數學
式而不受變數次數的影響,進而處理非線性的
限制條件。
此外,我們以速度和正確性為主要考量而
研製出一個完善的推理核心。它的目的當然是
在快速且正確地找出待解問題的答案。在CREDIT
的推理核心中,我們發展了一個新的資料結構
---限制衍生網路(Constraint derivation network)。我
們利用限制衍生網路來表示限制滿足問題,更
將它應用在不同的演算法以尋求限制滿足題的
解答。
我們也提供了一些CREDIT的應用範例,藉此
證明CREDIT的實用性。最後,我們亦提供CREDIT和
其他相關的套裝軟體作的比較數據。用以說明
CREDIT的優點。

We provide the development of a constraint reasoning
environment, CREDIT (Constraint Reasoning Environment Developed In
Taiwan-university). It is a package for solving constraint
satisfaction problems (CSP).
We define a special language to represent the model of various
CSPs. It is a humanlike representation to express the variables,
their domains and the relationships (constraints) among them. The
constraints can be represented either as formula or as oral
representations. The another advantage is it does not restricted by
higher order mathematical formula.
The particular data structure we used to record the
information of CSP model in memory is constraint derivation network.
It conquers not only the limitation of high degree but also
increases the number of constraints.
We also provide some experimental results of the CREDIT and
these results are compared to other famous constraint reasoning
systems.

[科資中心編號]RB8701-0192
[題    名]
自主型步行機器人的學習式控制程式之開發
Development of Learning Algorithms for an Autonomous Legged Walking
Robot.
[研 究 者 ]
張耀仁
Chang, Yau-Zeu
[機構名稱  ]長庚醫學暨工程學院機械工程系(CGMTMEE)
[經費來源  ]行政院國家科學委員會(NSC)
[補助金額  ]307000
[補助編號  ]NSC84-2212-E182-002
[研究開始日期]1994/08
[研究結束日期]1995/07
[頁 冊 數 ]51頁
[語    文]英文
[建檔日期  ]1998/10
[出版情形  ]完成日期:民國86年5月
[原件提供  ]行政院國家科學委員會科學技術資料中心
NSC84-2212-E182-002
[科資分類號 ]IF0303
[關 鍵 詞 ]
智慧型機器人;人工智慧;自主型機器人;人工生命;基因式學習法

Intelligent robot;Artificial intelligence;Autonomous robot;Artificial
life;Genetic learning

[摘    要]
本計畫的工作重點在於藉由改進R.A. Brooks所提出「基於行為模式的演算法」,去控制一具行動式機器人,以進行整合機器人技術和人工智慧的研究。由於構想中的機器人系統必需在無法預期且變動不定的真實環境中操作,從感知器獲得外界資訊到附諸行動之間要能夠在沒有人協助之下進行目標選定與行動規劃的推理。
我們使用基因式程式演化法於一六足機器人Genghis上,用以協調其步伐行為。為訓練其步行動作,作為回饋的根據是跌倒傾斜時,腹部著地的訊號。每一足各自獨立根據其他足的姿態去決定下一步行為。實驗結果顯示,某種經由簡化搜尋空間的方法可以有效的用基因式程式演化法找到有用的程式。

The effort of this project is to integrate research in robotics
and Artificial Intelligence by extending the Behavior-Based Programs
proposed by R.A. Brooks to control an actual mobile robot. Because
the proposed robotic system must operate in the real world which is
unpredictable, dynamic and uncertain, reasoning about goals and plans
without human assistance is essential for mapping perceptions onto
appropriate actions.
In this work we try to present a variation of the genetic
programming (GP) paradigm applied to a simple six-legged robot,
Genghis, learning to coordinate leg behaviors.
The only feedback is based on the state of a switch determining
whether the belly of the robot is touching the floor. Independent
behaviors on each leg monitor the activity of other behaviors and
correlate that activity, their own activity state, and the results
from the belly switch, as input to a local learning rule. The
correlation is used to learn under which conditions each behavior is
to become active.
Experimental results show that a specially designed approach,
where the search space has been much simplified, can generate
reasonable programs using genetic programming techniques.

[科資中心編號]RA8501-2032
[題    名]
分散式人工智慧系統發展工具之設計與實作
Design and Implementation of a Distributed Artificial Intelligence
System Development Tool.
[研 究 者 ]
何裕琨
HO,YU-KUEN
[機構名稱  ]成功大學電機工程系(NCKEELE)
[經費來源  ]行政院國家科學委員會(NSC)
[補助金額  ]327000
[補助編號  ]NSC84-2213-E006-010
[研究開始日期]1994/08
[研究結束日期]1995/07
[出版日期  ]1995/07
[頁 冊 數 ]54頁
[語    文]英文
[建檔日期  ]1996/10
[出版情形  ]研究報告,54頁,民國84年7月
Report, 54p., Jul. 1995
[原件提供  ]行政院國家科學委員會科學技術資料中心
NSC84-2213-E006-010
[科資分類號 ]IA0301
[關 鍵 詞 ]
分散式系統;人工智慧系統;解決問題;物件導
向;複層

Distributed system;Artificial intelligence system;Problem solving;
Object orientation;Multilayer

[摘    要]
本報告旨在分散式計算機網路系統環境
下,採物件導向之觀念(Object-oriented)以設計一分
散式協力解題殼體(Cooperative distributed problem
solving shell)。此系統運用分散式人工智慧
(Distributed artificial intelligence)技術,並結合物件
導向之觀念,對於分散式協力解題應用系統之
開發可提供一有效之發展環境。
由於每一個代理者(Agent)是以多層(Multi-
layer)方式建構而成的,因此可以降低設計的複
雜性且提供更多的彈性,並且將這些代理者安
排成遞迴式群組架構(Recurrent group structure)以方
便於執行環境之有效管理。除此之外,本報告
引用調停者(Mediator)之觀念以整合現有的系統,
以分散式白板代理者(Distributed whiteboard agent)作
為代理者間資訊分享的工具。
此一具物件導向觀念之分散式人工智慧系
統建構工具,包括應用系統架構描述(Architecture
description of application system)語言,負責工作的分
派與協調之動態工作推論機構(Reasoning mechanism
of dynamic task schedule),及基本通訊管理機構
(Management mechanism of essential communication)等相關
之軟體,均已設計完成。本系統經以實例加以
測試,對系統發展者而言可以很快地開發其特
定的協力式解題應用系統,而在執行的效率上
亦有甚為優異的表現。

In this report an Object-Oriented Cooperative Distributed
Problem Solving Shell (OOCDPSS) is presented. It is a convenient and
effective building tool for the cooperative multi-agent application
systems under distributed computer environment. By the Distributed
Artificial Intelligence (DAI) technique and the Object-Oriented (O-O)
concept, each agent is constructed by multi-layer structure to
reduce the design complexity and give more flexibility, and they are
arranged as a recurrent group structure for run-time effective
management. Additionally, the idea of mediator is adopted to
integrate new information systems with the existing inflexible
legacy system, and the distributed whiteboard agent is used for the
information interchange among the agents. This prototype system had
been completed and used as the test bed of our other laboratory
projects.


科資中心編號: RA8507-2088 
題名: 一個以角度參數為基礎的證據推理模式之探討 
Investigation of an Angular-Parameter-Based Evidential Reasoning Model. 
研究者: 鄧洪聲 
Don,Hon-Son 
機構名稱: 中興大學電機工程系(NCHSELE) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 269 仟元 
補助編號: NSC83-0408-E005-009 
研究日期: 1994/02 - 1995/01 
出版日期: 1995/01 
頁冊數: 36頁 
原件提供: 行政院國家科學委員會科學技術資料中心 NSC83-0408-E005-009 
科資分類號: IA0407 
關鍵詞: 人工智慧 ;專家系統 ;證據推理 ;相信函數 ;相依關係 ;證據結合 
Artificial Intelligence ;Expert System ;Evidential Reasoning ;Belief Function ;Dependency Relationship ;Belief Combination 
中文摘要: 我們提出一種新的證據推理模式。在這個模式中,每一個證據是以一個相信函數來表示,這個相信函數可以是分離式的,也可以是連續式的。我們考慮使用一個角座標的參數來表示兩個證據之間的的相依關係。我們在理論上分析這種模式的穩定性及適用性,並且探討在這種模式中各種證據推理的運算如何進行,這些運算包括證據之結合、證據之傳播、且/或運算、以及在一個規則中之傳播等。我們並且解決了一個存在已久的問題,也就是當有許多證據存在時,它們的結合次序會影響到最後的結果這個問題。我們做的一些探討顯示這個模式在解決上述問題上比現有的一些模式好很多。我們以實驗驗證,以對這個模式的特性有所瞭解。 
英文摘要: We propose a new evidential reasoning model in which a piece of evidence is represented by a belief function. The belief function can be in discrete or continuous form. We consider to use an angular parameter to represent the mutual dependency relationship between two pieces of evidence. We have theoretically analyzed the stability and feasibility of this model, and investigated how the various operations of evidential reasoning are performed. These operations include belief combination, belief propagation, AND/OR, and propagation within a rule, etc. We have also solved a long existing problem, i.e., the conflict resulting from the combination order when multiple evidence exist. Our investigation shows that the proposed model has great advantages over the conventional approaches in solving above problem. Some experiments have been conducted to demonstrate the properties of this model. 


[科資中心編號]RA8507-2071
[題    名]
限制條件推論系統之研製
Design of the Constraint Reasoning System.
[研 究 者 ]
許清琦
HSU,CHING-CHI
[機構名稱  ]台灣大學資訊工程研究所(NTUGCSIG)
[經費來源  ]行政院國家科學委員會(NSC)
[補助金額  ]366000
[補助編號  ]NSC83-0408-E002-020
[研究開始日期]1994/02
[研究結束日期]1995/01
[出版日期  ]1995/02
[頁 冊 數 ]84頁
[語    文]英文
[建檔日期  ]1997/01
[出版情形  ]研究報告,84頁,民國84年2月
Report, 84p., Feb. 1995
[原件提供  ]行政院國家科學委員會科學技術資料中心
NSC83-0408-E002-020
[科資分類號 ]IA0407
[關 鍵 詞 ]
推理;限制滿足問題;人工智慧;限制傳導;專家
系統

Reasoning;Constraint satisfaction problem;Artificial intelligence;
Constraint propagation;Expert system

[摘    要]
限制條件滿足問題(Constraint satisfaction
problems)在人工智慧中佔了非常重要的地位。在
許多情況下,CSPs可能解的範圍非常廣泛,以致於
一個有效率的搜尋演算法是很重要的。
Tree-based constraint propagation是一個解區間限
制條件滿足問題(ICSPs)的增強的方法。這個演
算法結合了傳導(Propagation)、區間計算(Interval
arithmetic)和樹狀結構(Tree structure)的技術。它可
以處理的限制問題不止是Hyvonen提出的"="關係
式子,更可以解">"、".gtoreq."、"<"、".ltoreq."和
".neq."等關係式的限制。
我們建立了一個有效的推理系統,
CREDIT(Constraint reasoning environment developed in Taiwan-
university)。這個系統分成兩個部分:第一部分是
一個前置處理器(Preprocessor),它允許一個不熟悉
電腦語言的使用者將他的問題寫成限制程式,
同時並將這個簡單的程式轉換成C程式碼。第
二部分是核心部分,在這裡包含了根據各種不
同的方法所寫成的子程式(Subroutine),而這些子
程式可以用來解不同型態的限制問題。

Constraint Satisfaction Problems play major roles in the field
of artificial intelligence since many problems can be formulated as
CSPs. The solution space (possible solutions), in most cases, is
large and the relations in CSPs are diverse that an efficient search
method that can deal with more general relations in CSPs is needed.
An enhanced strategy, called tree-based constraint propagation,
is proposed to solve Interval Constraint Satisfaction Problems. This
method combining the propagation, interval arithmetic and tree
techniques and can deal with not only the relation of "=" that
proposed by Hyvonen, but also the relations of ">", ".gtoreq.", "<",
".ltoreq." and ".neq.".
We construct a reasoning system, called CREDIT (Constraint
Reasoning Environment Developed In Taiwan-university), with the
adoption of the strategy. Our system can be divided into two parts:
the first part (a preprocessor), which translate simple constraint
language to C code, allows users unfamiliar with computer languages
to express their problems in easy-understood constraint languages.
The system kernel is the second part, which is composed by a set of
subroutines (called library). Different approaches that solve
different constraint relations are encapsulated as subroutines to
facilitate problem solving.

[科資中心編號]RB8707-0491
[題    名]
收縮型自動推論方法
Contraction-based Automated Deduction Methods
[研 究 者 ]
項潔
Hsiang, Jieh
[機構名稱  ]台灣大學資訊工程研究所(NTUGCSIG)
[經費來源  ]行政院國家科學委員會(NSC)
[補助金額  ]177000
[補助編號  ]NSC83-0408-E002-012-T
[研究開始日期]1993/08
[研究結束日期]1994/07
[頁 冊 數 ]53頁
[語    文]英文
[建檔日期  ]1999/10
[出版情形  ]完成日期:民國85年1月
[原件提供  ]行政院國家科學委員會科學技術資料中心
NSC83-0408-E002-012-T
[科資分類號 ]IA0407
[關 鍵 詞 ]
自動推理;推論法則;決策程序;人工智慧

Automated deduction;Inference rule;Decision procedure;Artificial
intelligence (AI)

[摘    要]
本報告包括國科會計畫「收縮型自動推理方法」(NSC-83-0408-E002-012T)期末報告研究之部分結果。本報告之主要貢獻是提出一個以收縮型Completion procedure做為Semi-decision procedures的理論基礎。Completion procedures為Knuth-Bendix於1969年提出,其目的為做為Generate等式系統(Equational theory)的Decision procedures的方法。我們提出一個新的架構,在其下Completion procedures亦可用來做自動推理的Semi-decision procedures。其最大的不同在為Semi-decision procedures是Traget-oriented,而且具有邏輯上的完備性。我們將推理法則(Inference rules)分成擴展型(Expansion)及收縮型(Contraction)。後者在產生新的推理同時也去除掉一些對證明無用的(Redundant)data。對搜尋方法我們提出一個Fairness的觀念。在我們的架構下可以解釋並延伸所有的Completion procedures成Target-oriented semi-decision procedures。

This report describes some of the results obtained under the
sponsorship of the National Science Council grant NSC83-0408-E002-
012T the title of the proposal is "Contraction-based automated
deduction". The main contribution of this report is to give an
outline of a logical foundation of semidecision theorem proving
procedures based on completion procedures.
Completion procedures, originated from the seminal work of
Knuth and Bendix, are well-known as procedures for generating
confluent rewrite systems, i.e. decision procedures for equational
theories. We present a new abstract framework for the utilization of
completion procedures as semidecision procedures for theorem proving.
The key idea in our approach is that a semidecision process should be
target-oriented. i.e. keep into account the target theorem to be
proved. For the inference rules of a completion procedure, we present
target-oriented schemes of contraction inference rules, i.e.
inference rules that delete sentences which are redundant for proving
the target. For the search plan, we give a target-oriented,
definition of fairness, according to which not all critical pairs
need to be considered. We prove that our notion of fairness, together
with the refutational completeness of the inference rules, is
sufficient for a completion procedure to be a semidecision procedure.
By relaxing the requirement of considering all critical pairs, our
target-oriented framework should be more suitable for designing
efficient procedures for theorem proving. The generation of decision
procedures is included as a special side-effect and all the results
of the classical approach to completion are re-obtained in our
framework. The application of completion to disprove inductive
conjectures, i.e. the so called inductionless induction method, is
also covered as a semidecision process. Finally, we present,
according to our framework, some equational completion procedures
based on Unfailing Knuth-Bendix completion.


[科資中心編號]RA8410-2090
[題    名]
機器學習理論在自然語言習得的研究(II)
A Study on Machine Learning Theories and Natural Language
Acquisition. (II)
[研 究 者 ]
蘇豐文
SOO,VON-WUN
[機構名稱  ]清華大學資訊科學系(NTHUCOS)
[經費來源  ]行政院國家科學委員會(NSC)
[補助金額  ]817000
[補助編號  ]NSC82-0408-E007-029
[研究開始日期]1992/08
[研究結束日期]1993/07
[出版日期  ]1994/02
[頁 冊 數 ]298頁
[語    文]英文
[建檔日期  ]1996/06
[出版情形  ]研究報告,298頁,民國83年2月
Report, 298p., Feb. 1994
[科資分類號 ]IA0407;IA0405
[關 鍵 詞 ]
機器學習;人工智慧;自然語言習得;類神經網
路;歸納學習法

Machine learning;Artificial intelligence;Natural language
acquisition;Artificial neural network;Empirical generalization

[摘    要]
本研究的主要目的是研究人工智慧領域裡
機器學習理論與自然語言習得的關係。一般而
言,機器學習理論可概分為以解釋為主的學習
法(Explanation-based learning,EBL)與以相似性為主的
學習法(Similarity-based learning,SBL)。而SBL又可由
多種方法來達成如:觀念堆集法(Conceptual
clustering)、以案例為主的學習法(Case-based
learning)、類神經網路模型。如何結合EBL與SBL來
達成有效的學習,便是我們研究的主題。我們
將尋求適當的機器學習理論模型,應用到自然
語言的習得中。理想的自然語言習得模型理應
與習得的目標語無關,我們將從英語與漢語為
學習的目標語,以評估機器學習模型的效益與
可行性。我們將往年在自然語言語法剖析與習
得方面的研究成果與經驗加以擴充並深入探討
語意與篇章訊息方面的處理與習得等問題。主
要的研究分兩方面來進行: (1)研究如何將語言
學理論如GPSG作為EBL的領域理論來幫助語言的
習得? (2)如何用反複式類神經網路來學習剖析
具有歧義及不合格的句子? (3)如何用CBL來歸納
漢語前指現象的規律?

Domain theory plays an important role in an Explanation-Based
Learning (EBL) system. Its adequacy determines the portability of
the learning model and the learnability of the target knowledge
pieces. In this paper, we propose a new natural language acquisition
model (called EBNLA) based on EBL by incorporating universal
linguistic principles as innate domain theory. The domain theory
consists of two parts: the static part and the dynamic part. The
static part, which is assumed to be innate to the model, includes
the theta-theory in Government-Binding theory (GB) and the universal
feature instantiation principles proposed in the Generalized Phrase
Structure Grammar (GPSG) formalism. The dynamic part contains
Context-Free grammar rules as well as syntactic and thematic
features of lexicons. In parsing (problem solving), both parts work
together to parse input sentences. As parsing fails, learning is
triggered to enrich and generalize the dynamic part by obeying the
principles in the static part. By introducing EBL and the universal
linguistic principles, the learnability of knowledge pieces and the
portability of the model can be improved.

科資中心編號: RA8507-2079 
題名: 邏輯函數類神經網路
The Logic Function Neural Network. 
研究者: 曾怜玉 
Tseng,Lin-Yu 
機構名稱: 中興大學應用數學研究所(NCHUAPMG) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 327 仟元 
補助編號: NSC83-0408-E005-005 
研究日期: 1993/08 - 1994/07 
出版日期: 1994/08 
頁冊數: 40頁 
原件提供: 行政院國家科學委員會科學技術資料中心 NSC83-0408-E005-005 
科資分類號: IA0407 
關鍵詞: 類神經網路 ;人工智慧 ;函數鏈路神經網 ;學習法則 ;邏輯函數神經網 
Artificial Neural Network ;Artificial Intelligence ;Functional-Link Neural Network ;Learning Rule ;Logic Function Neural Network 
中文摘要: 所有前饋式類神經網路皆適合在監督學習後當做模式分類器。在前饋式類神經網路中,後向傳播學習法則是一種功能很強,且已經成功應用到各種不同的問題上。然而,因後向傳播學習法則是一種「黑箱」學習法,故仍然存在某些缺點,特別是學習時間往往比較長,當學習之樣本數量太大時,學習成功率不是很理想,以及隱藏單元數目無法預知等缺點。因此,Pao提出函數鏈路類神經網路(Functional-link neural network)來減少隱藏層和增快學習速度。本計畫中,我們提出利用網路分割特性和交換邏輯理論之方法,設計一「邏輯函數類神經網路」以增進前饋式類神經網路之學習速度。由我們的實驗結果可知,前饋式類神經網路不但可擺脫傳統之盲目學習,而且使用我們提出之邏輯函數類神經網路還可獲得下列優點:網路變小、學習成功率提高、以及學習速度變快。 
英文摘要: Feedforward networks are good pattern classifiers, and can be used with supervised learning. Among the feedforward network learning rules, the backpropagation learning rule is a powerful one which has been successfully applied to a variety of problems. However, the process of the backpropagation learning is somewhat a"blind learning". It suffers from several limitations, in particular, the learning speed is often unacceptably low, the recognition rate may be low if there are a large number of input patterns, and the number of hidden units is not known in advance. Therefore, functional- link neural network was proposed by Pao to eliminate hidden layers and improve the speed of learning. However, the drawback of such a network is the combinatorial explosion of the high-order terms. In this project, we propose logic function neural network (LFNN). An input-output relation with multiple output nodes is first divided into several input-output relations, each with exactly single output node, then a logic function neural network is designed by applying logic design theory for each such input-output relation to help the network learning so as to improve the performance. Our experimental results show that logic function neural network can learn faster because it does not learn blindly. The advantages of the proposed neural network are: the network size becomes smaller, the recognition rate is higher, and the learning speed is faster as well. 

科資中心編號: RA8307-2019 
題名: 網路管理專家系統知識擷取工具之研製(II) 
Development of a Knowledge Acquisition System for Network Management Expert System. (II) 
研究者: 曾憲雄 ; 黃國禎 
Tseng,Shian-Shyong ; Hwang,Gwo-Jen 
機構名稱: 交通大學資訊科學研究所(NCTUINSG) 
經費來源: 交通部電信總局電信研究所(TCLG) 
補助編號: TL-82-3107 
研究日期: 1992/11 - 1993/10 
出版日期: 1993/12 
頁冊數: 60 
科資分類號: IA;IA0407;IA0202 
關鍵詞: 網路管理 ;知識擷取 ;網路模擬系統 ;專家系統 ;人工智慧 
Network Management ;Knowledge Acquisition ;Network Simulation System ;Expert System ;Artificial Intelligence 
中文摘要: 近年來由於人工智慧科技的進步,將人工 智慧的理念與技術應用於管理的例子已履見不 鮮,其中最受歡迎的,莫過於專家系統理論之應 用。同時,由於網路上的傳輸規格已進展到足 以使個別計算機都可取得整體網路之狀態資 訊,而使網路管理之自動化更為可行。因此,利 用專家系統之理論與技術,發展出有效管理網 路的方法與系統,已成為未來必然趨勢,而在發 展專家系統的過程中,知識擷取是最艱鉅且最 耗時的工作。因此,在交通部電信研究所的支 持與協助下,我們發展了一套知識擷取系統。 這套系統是以Repertory-Grid及EMCUD等相關文獻內 容作為理論基礎,同時在知識擷取程序完成之 後,可以產生著名專家系統建構軟體CLIPS之知識 庫。此外,本系統並提供專家及知識工程師不 同的介面及知識編輯器,使得專家系統的建立 更為精確便利。我們也計畫在未來擴充這套系 統,使其在應用上與操作上都更為完善。 
英文摘要: In recent years, the requirements of data communications and message transfers grow in a extremely surprising manner, and hence the development of computer network has become one of the most important issues today. However, as computer networks grow fast, it becomes a difficult and important problem to manage the network systems. As expert system approach is known to be one of the most popular techniques to solve the management problems. In our previous project, we try to cope with the network management problems with this approach. We proposed a knowledge acquisition model for network management, since knowledge acquisition is known to be a critical model in building expert systems, and we believe that as long as the bottleneck is solved, the development of network management expert systems can be well done. In this project, we plan to develop a knowledge acquisition for network management based upon the model previously proposed. The knowledge acquisition will be able to interview with experts and produce knowledge bases for a well-known expert system shell, CLIPS. We shall also provide an interface for knowledge engineers to modify the knowledge bases. By employing the knowledge acquisition system, we plan to build an expert system for network management. The resulted expert system will be able to advise some parameters concerning network configuration. The parameters will be the input data of a network simulation system, LANNET II, to simulate the behaviors of the configured network and to evaluate the performance of the expert system. 


科資中心編號: RA8410-2090 
題名: 機器學習理論在自然語言習得的研究(II) 
A Study on Machine Learning Theories and Natural Language Acquisition. (II) 
研究者: 蘇豐文 
Soo,Von-Wun 
機構名稱: 清華大學資訊科學系(NTHUCOS) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 817 仟元 
補助編號: NSC82-0408-E007-029 
研究日期: 1992/08 - 1993/07 
出版日期: 1994/02 
頁冊數: 298頁 
科資分類號: IA0407;IA0405 
關鍵詞: 機器學習 ;人工智慧 ;自然語言習得 ;類神經網路 ;歸納學習法 
Machine Learning ;Artificial Intelligence ;Natural Language Acquisition ;Artificial Neural Network ;Empirical Generalization 
中文摘要: 本研究的主要目的是研究人工智慧領域裡機器學習理論與自然語言習得的關係。一般而言,機器學習理論可概分為以解釋為主的學習法(Explanation-based learning,EBL)與以相似性為主的學習法(Similarity-based learning, SBL)。而SBL又可由多種方法來達成如:觀念堆集法(Conceptual clustering)、以案例為主的學習法(Case-based learning)、類神經網路模型。如何結合EBL與SBL來達成有效的學習,便是我們研究的主題。我們將尋求適當的機器學習理論模型,應用到自然語言的習得中。理想的自然語言習得模型理應與習得的目標語無關,我們將從英語與漢語為學習的目標語,以評估機器學習模型的效益與可行性。我們將往年在自然語言語法剖析與習得方面的研究成果與經驗加以擴充並深入探討語意與篇章訊息方面的處理與習得等問題。主要的研究分兩方面來進行: (1)研究如何將語言學理論如GPSG作為EBL的領域理論來幫助語言的習得? (2)如何用反複式類神經網路來學習剖析具有歧義及不合格的句子? (3)如何用CBL來歸納漢語前指現象的規律? 
英文摘要: Domain theory plays an important role in an Explanation-Based Learning (EBL) system. Its adequacy determines the portability of the learning model and the learnability of the target knowledge pieces. In this paper, we propose a new natural language acquisition model (called EBNLA) based on EBL by incorporating universal linguistic principles as innate domain theory. The domain theory consists of two parts: the static part and the dynamic part. The static part, which is assumed to be innate to the model, includes the theta-theory in Government-Binding theory (GB) and the universal feature instantiation principles proposed in the Generalized Phrase Structure Grammar (GPSG) formalism. The dynamic part contains Context-Free grammar rules as well as syntactic and thematic features of lexicons. In parsing (problem solving), both parts work together to parse input sentences. As parsing fails, learning is triggered to enrich and generalize the dynamic part by obeying the principles in the static part. By introducing EBL and the universal linguistic principles, the learnability of knowledge pieces and the portability of the model can be improved. 


科資中心編號: RA8410-2027 
題名: 加速神經網路學習之研究(I) 
A Study of Speeding up Learning of Neural Networks. (I) 
研究者: 洪宗貝 ; 許文龍 ; 曾憲雄 
Hong,Tzung-Pei ; Shu,Wen-Lung ; Tseng,Shian-Shyong 
機構名稱: 中華工學院資訊工程系(CHPICOS) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 393 仟元 
補助編號: NSC82-0408-E216-004 
研究日期: 1992/08 - 1993/07 
出版日期: 1993/07 
頁冊數: 80頁 
科資分類號: IA0407 
關鍵詞: 類神經網路 ;人工智慧 ;機器學習 
Artificial Neural Network ;Artificial Intelligence ;Machine Learning 
中文摘要: 近年來,神經網路已可成功地被應用在解決許多困難的問題,如辨認、診斷、控制等問題上。但一個神經網路執行效率的好壞,則由其上的權重係數決定。在過去,雖然有不少自動學習出此權重係數的演算法已被提出,但當所用來學習的訓練用例子數目很多時,則常須很久的學習時間。在此計畫中,我們嘗試利用一種非常容易建構的平行模式, Single-channel broadcast communication model,來加速神經網路的學習過程。在此計畫的第一年,我們將從最簡單的神經網路模型-Perceptron開始探討。我們首先提出在Single-channel broadcast communication model上利用N個元素(N為所要處理的訓練用例子數目)的平行 Perceptron學習演算法。在不同的假設之下,我們可分析出此平行演算法平均加速程度介於O(log N)至O(N)之間。實驗結果顯示平均加速程約為 O(N/sup 0.91//log N)。此外,我們亦討論了當所能使用的處理器數目小於訓練例子數目的處理方式。我們並提出了一種新平行學習演算法以解決此情形。理論分析及實驗結果均顯示當所使用的處理器數目r遠小於訓練用例子N時,平均加速程度可達O(r)的效果,因此我們所提出的平行學習演算法相當適用於所要處理的訓練例子相當多的情況。最後,我們並提出一種較具有彈性的機率式學習演算法,使得此演算法可在我們所允許的錯誤率容忍範圍內即停止執行,以進一步節省所須的學習時間,這對於雜訊存在的學習環境下之學習非常適合。此方法使得執行時間和結果準確性之間可根據問題的需要而達成一種折衷。最後之實驗結果並驗證我們的方法可以在犧牲少許的準確性下,節省相當多的執行時間。 
英文摘要: Neural network models are drawing interesting as useful tools for A.I. They have successfully been applied to many difficult problems such as recognition, diagnosis, control, and so on. The behavior of a neural network mainly depends on the weights of the connections between its cells. If the values of these weights are inappropriately assigned, the performance of the neural network will be greatly lowered. Assigning appropriate values to the weights is then very important. In the past, many learning methods were proposed for automatically achieving values of weights from a set of training instances. The needed learning time is, however, quite long especially when the number of training instances is large. Speeding up the efficiency of learning in the neural network is then the main purpose in this project. In the first year of this project, we have adopted an easily implemented parallel model, single-channel broadcast communication model, for speeding up learning of single-layer perceptron. Given n training instances and a processors, theoretical average speed-up has been shown to range from O(log n) to O(n) under a variety of assumptions. Empirical results have also shown average speed-up approximately being O(n/sup 0.91//log n). Next, we discuss the situation where the number of processors is less than the number of training instances. A parallel perceptron learning algorithm on the single-channel broadcast communication model with r processors (r<n) has then been proposed. Given r processors, the average speed-up has been shown to be near r for a large (compared with r) training set. The single-channel broadcast communication model then provides another good candidate for speeding up learning of neural networks especially when the number of training instances is quite large. Besides, we have also designed a probabilistic perceptron learning algorithm which can terminate when the error ratio of a derived weight vector is within the admitted tolerance. The needed learning time can then be further saved. In the noisy learning environment, allowing a tolerance of error is reasonable. The proposed learning algorithm is then very suitable for managing noisy training instances. The idea used in the parallel perceptron learning algorithms and the probabilistic perceptron learning algorithm will be extended to other neural learning models (such as the back-propagation learning model) in the forthcoming next year of this project. We hope the results in this project can provide a feasible solution to effectively speed up the learning of neural networks. 

科資中心編號: RA81-2025 
題名: 專家系統語言的研究 
On Study of Expert System Languages. 
研究者: 李錫智 
Lee,Shie-Jue 
機構名稱: 中山大學電機工程研究所(NSYSELEG) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 378 仟元 
補助編號: NSC81-0408-E110-02 
研究日期: 1991/08 - 1992/07 
出版日期: 1992/08 
頁冊數: 116 
科資分類號: IA;IA0407 
關鍵詞: 專家系統 ;推論方式 ;人工智慧 ;編譯器 ;語法 ;語意學 
Expert System ;Inference Method ;Artificial Intelligence ;Compiler ;Syntax ;Semantics 
中文摘要: 專家系統是人工智慧技術應用中最成功的 一環,它在醫療、診斷、計畫、排程方面都有 成功的例子。目前發展專家系統的語言(如 Prolog、LISP、CLIPS等)大致上可以區分為三類:基 於函數式的、基於邏輯表示式的、與基於法則 式的。他們在語意語法、控制結構與發展工具 上各有其優缺點。本計畫的目的是想發展一個 新的專家系統語言。這個語言同時具備了函數 式、邏輯式與法則式表示法的優點。本篇報告 記錄了這項計畫在第一年的研究成果。我們探 討了一些常用的推論方式,並比較其優缺點。 其次,我們研究了幾種常用的專家系統語言 (Prolog、LISP、CLIPS、OPS5),對其表示方式、控制 策略上作了比一番介紹與比較,以了解這些專 家系統語言的不便與所受到的限制。我們發現 雖然規則網路使專家系統得以實用化,但是它 尚存有知識不一致的問題。我們提出一個改良 規則網路的方案。這項方案包括法則動態增減與資料再配對的技術。因此,使得學習能力能 夠在專家系統中方便快速地被整合。目前,這 個專家系統語言的語意語法、控制策略、偵錯 及發展工具是以C-Prolog來模擬。由測試中發現, 這項改良方案確實可行,可以成功地解決一些 其他專家系統語言辦不到的問題。 
英文摘要: Expert systems are increasingly important to industry, business, and defense affairs. Expert system languages are important to the development of expert systems. An expert system language can be divided into three parts: knowledge representation, control strategies, and debugging and developing tools. In this report, we survey some reasoning methods and some current expert system languages on their knowledge representation styles, control strategies, and advantages and disadvantages. The rule network which performs pattern matching efficiently is discussed and improved. We propose a new expert system language which has logic-based, function-based, and rule-based flavors. The control strategy is improved so that learning can be performed flexibly in expert systems developed in this language. A system that simulates our idea is constructed. The expert system language is proved to be effective by solving some problems successfully which can not be solved easily by other expert system languages. The simulation and test results are presented. 



科資中心編號: RA8510-2086 
題名: 建構能處理模糊輸入資料的類神經網路模式(I) 
Building a Neural Network Model with Fuzzy Inputs. (I) 
研究者: 李漢銘 
Lee,Hann-Ming 
機構名稱: 台灣工業技術學院電子工程技術系(NTITENT) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 327 仟元 
補助編號: NSC83-0408-E011-002 
研究日期: 1993/08 - 1994/07 
出版日期: 1994/09 
頁冊數: 57頁 
原件提供: 行政院國家科學委員會科學技術資料中心 NSC83-0408-E011-002 
科資分類號: IA0407 
關鍵詞: 人工智慧 ;類神經網路 ;模糊輸入 ;模糊計算 ;學習法則 ;語言學 
Artificial Intelligence ;Artificial Neural Network ;Fuzzy Input ;Fuzzy Computing ;Learning Rule ;Linguistics 
中文摘要: 申請人在上個研究計畫「具模糊計算能力之類神經網路推理系統」中將模糊計算能力加入類神經網路中。本計畫旨在接續此一研究,期望能真正同時處理模糊及精確表示之資料,以提供處理Uinguistic及Ambiguous資訊之良方。除此之外,利用模糊集合理論來提高類神經網路的學習速度,及對網路運作功能及特性提供 Uinguistic描述等方面的研究亦是我們研究的目標。本計畫擬於兩年內完成,本年度首先探討模糊理論中對於模糊資料各種表示法的運算特性;然後針對類神經網路處理模糊輸入加以分析;同時我們將對學習法則所應採用的方式作深入分析。下一年度則進行整合設計、製作與模擬的工作。 
英文摘要: In our last project, "Building Connectionist Inference System with Fuzzy Computing Ability", we have successfully incorporated fuzzy computing into the connectionist inference system. This project will continue the work by handling both fuzzy and crisp inputs to provide the processing of linguistic and ambiguous information. In addition, using fuzzy set theory to speed up the learning speed of neural networks and providing linguistic description to the network operation and characteristics are also considered in this project. In the first year of this two year project we will investigate various fuzzy inputs representations and their corresponding operation. Also we will consider the handling of neural network models with fuzzy input. In addition, we will design the appropriate learning algorithm for proposed model. In next year, integrating design, implementation and simulation will be done. 


[科資中心編號]RA81-2668
[題    名]
以演進為基礎的學習自動機
Evolution-Based Learning Automata and Algorithms.
[研 究 者 ]
王勝德
WANG,SHENG-DE
[機構名稱  ]台灣大學電機工程研究所(NTUGELEG)
[經費來源  ]行政院國家科學委員會(NSC)
[補助金額  ]301000
[補助編號  ]NSC80-0408-E002-04
[研究開始日期]1990/08
[研究結束日期]1991/07
[出版日期  ]1991/09
[頁 冊 數 ]80
[語    文]英文
[建檔日期  ]1994/05
[科資分類號 ]IE;IE03;IA0407
[關 鍵 詞 ]
學習自動機;人工智慧;演進演算法; 隨機自動機

Learning automata;Artificial intelligence;Evolutionary algorithm; Stochartic
automata

[摘    要]
演進程序或學習乃是以隨機的方式改變系 統的結構以
使系統從環境獲得不斷改善的報 償,本文研究兩種型式的以
演進為基礎的學習 演算法:(1)具時變學習增益的隨機學習自
動機, (2)以遺傳學習演算法為基礎的分類器系統,兩 者都是
透過與環境交談漸進改善系統性能。
分類器系統是一種次符號層次之學習模 型,其目的在於
從不定性環境中歸納出類似刺 激-反應型式的規則。其本身
是一種以規則為 基礎,具有高並行度的產生系統,輔以桶隊演
算 法來評估系統中規則之優劣,並以遺傳演算法 作為學習規
則的主要機制。
本文對於分類器系統的學習模型提出了多 項改良。文
中所發表的後向擴散演算法,對於 迴授資訊有較佳的利用率,
因而縮短了評估規 則優劣的程序。此外我們發展了限制刪
除及限 制產生的技術,用以控制世代變遷時分類器的 繁衍。
本文對於隨機學習自動機的學習模型提出 具時變學習
增益的改良,此時變學習增益是以 兩個參數來控制使增益由
小增至一高限後再隨 時間降低。
文中亦包含了相關的實驗報告,以做為本 文對於現有技
術改良之佐證。

Evolutionary processes or learning involve random changes in the
structure of a system which can lead to adaptation such that the reward gained from the environment are maximized. In this technical report, two types of evolution-based learning are examined: (1) Stochastic learning automata which maintain an estimation of environment's characteristics; (2) Classifier systems with genetic learning algorithms. Both models of learning improve their performances through interactions with the underlying environment.
Classifier systems are intended as a general approach to the induction
of need-serving behavior in uncertain environments. It is a rule-based,
message-passing, highly parallel sub-symbolic production system with bucket brigade algorithm as its credit assignment mechanism and genetic algorithm as its primary rule discovery mechanism. Following a brief introduction to classifier systems, various refinements for the learning model of classifier systems are presented, including the back diffusion algorithm, which makes better use of available feedback to speed up the credit assignment process; the restricted deletion and the restricted production of classifiers, which help to control the proliferation of classifiers during population evolution.
For the stochastic learning automata, we propose using a time-varying
gain together with the estimator approach in the learning algorithm to speed up the convergence. The proposed time-varying gain is controlled by two parameters which make the gain increase to a certain value initially and then decrease to zero as time proceeds. The simulation results show the superiority of the proposed learning scheme. We believe that this model can better fit a practical application area in this field.


科資中心編號: RA78-2028 
題名: 具Meta能力之專家系統建構工具
An Expert System Building Tool with Meta Ability. 
研究者: 何裕琨 ; 陳敬 
Ho,Yu-Kuen ; Chen,Jing 
機構名稱: 成功大學電機工程研究所(NCKUELEG) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 302 仟元 
補助編號: NSC78-0408-E006-01 
研究日期: 1988/08 - 1989/07 
出版日期: 1989/07 
頁冊數: 110 
原件提供: 行政院國家科學委員會(NSC), NSC78-0408-E006-001, 2片, 118幅 
科資分類號: IA;IA0407 
關鍵詞: 人工智慧 ;專家系統 ;專家系統建構工具 ;Meta問題 
Artificial Intelligence ;Expert System ;Expert System Building Tool ;Meta Problem 
中文摘要: 本報告旨在一專家系統建構工具中(Expert System Building Tool)設計一能使用中間-知識(Meta-knowledge)之推論機引擎(Inference Engine),此推論機能採用使用者預設之中間-知識以增進系統推論上之效率.由於推論上之需要,本報告首先設計出一種中間-知識的表示法,依據此種中間-知識,系統將在知識的推理過程中,選擇較為重要的法則優先予以推論,且忽略不相關的法則,使系統於知識的推論上,得到較高的效率.此外,在推論機的製作上,則在傳統的逆推式(Backward chaining)的推理架構中,加上類似前推式(Forward chaining)的功能,使得在推論的過程中,當某一屬性(attribute)的值(value)得到時,能預先將此值填充至知識庫具有相同屬性的法則中,使得在下一次的推論中,能夠減少知識庫的搜尋次數,因而提昇其推論的效率.而由於在知識均採用屬性-值(attribute-value)之結構,所以,在推論機的設計上,則採用以屬性為導向的推論設計,使得相同屬性的資料s能集中在一起,如此,系統在作知識之推論時,若某一法則的屬性-值不符,系統可立刻至相關於此屬性的另一法則去作推論,如此,可以免去對法則的搜尋時間,而對系統的效率之增進有甚大之助益.此一具有中間(Meta)能力之專家系統建構工具,包括知識表示法,可以處理中間-知識(Meta knowledge)的推論機,及以屬性為導向之推論設計等相關之軟體,均已設計完成.對於知識的推理過程及法則的搜尋效率上,經以實例加以測試與比較,證明其較其他專家系統建構工具有較高之推理效率. 
英文摘要: In this report, an inference engine using meta-rules is designed for an Expert System Building Tool. This inference engine using a traditional backward chaining embedded with forward-chaining-like control strategy is more efficiency than the other systems. For using the attribute oriented knowledge structure, the forward-chaining- like property can be seen at the attribute-value decision process. The meta- knowledge which represented as the meta-rule is used as the meta-level control knowledge. This meta-knowledge can provide the selection of paths of reasoning to improve the efficiency of reasoning. The software of this Expert System Building Tool, including the knowledge acquisition system, the inference engine, and the knowledge-data base management system, have been designed and implemented . By testing with some examples, when the meta-knowledge is added, the whole system is more efficiency than the non meta-knowledge condition. 

科資中心編號: RB9104-0211 
題名: MP3音樂資料庫系統的設計與實作(I) 
The Design and Implementation of an MP3 Music Database System 
研究者: 劉志俊 
Liu, Chih-Chin 
機構名稱: 中華大學資訊工程學系 
經費來源: 行政院國家科學委員會 
補助金額: 新台幣 430 仟元 
補助編號: NSC89-2218-E216-006 
研究日期: 2000/08 - 2001/07 
出版日期: 2001/10 
頁冊數: 5頁 
原件提供: 行政院國家科學委員會科學技術資料中心, NSC89-2218-E216-006 
科資分類號: IA0404;IA0406;IA0411 
關鍵詞: 音樂資料庫系統 ;多媒體內涵式檢索 ;MP3數位音樂檔案格式 ;隨選音樂 ;音樂特徵抽取 ;動態影像專家群 
Music Database System ;Multimedia Content Based Retrieval ;Mpeg Audio Layer 3 (Mp3) ;Music On Demand ;Music Feature Extraction ;Moving Picture Experts Group (Mpeg) 
中文摘要: 隨著多媒體應用系統與網際網路的快速興起,使用者可以透過網路方便地擷取到包含影像、聲音、視訊、動畫等多媒體資料。為了讓使用者能方便地在大量的多媒體資料中,必須提供使用者以一定的查詢條件來過濾出所想要的多媒體資料並提供更有效的多媒體資料儲存管理功能。因此多媒體資料庫的相關研究於是成為學術界與產業界研究的焦點之一。而隨著新一代多媒體標準如JPEG2000, MPEG4,與MPEG7 等標準的制定多媒體內涵式查詢(Multimedia Content-basedRetrieval)成為目前極其重要的研究課題。由於MP3(MPEG-1 Audio Layer 3)格式的音訊資料所具有的高壓縮比和近乎CD 的音樂品質之特性,大大的增加了MP3 音訊資料在網際網路上的應用。本篇論文試者對MP3 的內涵分析,找出MP3的特徵值並對其做索引,並根據MP3 特徵值建立MP3 資料庫,最後可讓使用者輸入查詢範例,根據使用者查詢範例和MP3 資料庫裡的資料作相似性的比對,找出使用者想要的MP3 物件出來。本論文所提出的方法也可以應用在隨選音樂(MusicOn-Demand)、MP3 搜尋引擎以及網路點歌等。 
英文摘要: In recent years, the searching and indexing techniques for multimedia data are getting more attention in the area of multimedia databases. As many research works were done on the content-based retrieval of image and video data, less attention was received to the content-based retrieval of audio data. In this paper, we propose an approach to retrieve MP3 music objects based on their content. In our approach, the coefficients extracting from the output of the polyphase filters are used to compute the MP3 features for indexing the MP3 objects. We also propose an MP3 similarity measuring function to provide users the ability to approximately retrieve the desired MP3 objects. Experiments are performed and analyzed to show the efficiency and the effectiveness of the proposed method. 

科資中心編號: RN9102-0287 
題名: 多模態音樂檔案快速關鍵旋律自動擷取及其應用
Fast Key Melody Extraction for Multi-Modal Music Files and Its Application 
研究者: 曾元顯 
機構名稱: 輔仁大學圖書資訊系 
經費來源: 行政院國家科學委員會 
補助編號: NSC89-2413-H030-016 
研究日期: 2000/08 - 2001/07 
出版日期: 2001/10 
頁冊數: 4頁 
原件提供: 行政院國家科學委員會科學技術資料中心, NSC89-2413-H030-016 
科資分類號: PA0210 
關鍵詞: 關鍵旋律擷取 ;音樂檔案 ;快速演算法 ;音樂內容查詢 ;音樂內容導覽 ;雜湊函數 ;MIDI檔案 
Key Melody Extraction ;Music File ;Fast Algorithm ;Content-Based Music Retrieval ;Content-Based Navigation ;Hash Function 
中文摘要: 關鍵旋律(或主題音樂)的擷取與應用對提升音樂內容檢索的成效非常重要。如同關鍵詞的功能一樣,關鍵旋律也可以運用於音樂數位圖書館或音樂超媒體系統裡,提供內容式導覽、摘要或視覺化的功能。我們發展出來的關鍵旋律自動擷取方法,是一種擷取最大可重疊重複片段的方法,其計算複雜度的理論最佳值為 O(LN)。這個結果是運用到雜湊函數且假設雜湊函數的平均存取時間為理論的最佳值O(1)。然而,當處理的資料長度越長時,要實作一個雜湊函數具備理論上最佳的計算時間將變得相當困難。本計畫即在既有的基礎上,針對不同模態的數位錄音檔案,進行關鍵旋律快速擷取的研究,並應用於音樂內容檢索與導覽上,以支援音樂資料庫或音樂數位圖書館多種檔案格式的超媒體應用。 
英文摘要: Previous researches show that key melodies (or themes) play an important role in improving the performance of content-based music retrieval. Similar to the roles of keywords, key melodies can further provide content-based navigation, abstracting, and visualization functions in digital music library systems or hypermedia music systems. We have developed a key melody extraction algorithm for MID files. This algorithm is able to extract maximum overlapped repeated patterns with time complexity O(LN), in theory. This result is based on the use of hash functions and based on the assumption that the access time of hash values is constant theoretically. However, as the length of data getting longer, implementing a hash function with theoretic optimal time complexity becomes very difficult. This project will base on the past results to develop practical fast key melody extraction method for MIDI, MP3, and other multi-model music files. The fast extraction method will apply to content- based retrieval and navigation to support digital music library and hypermedia music systems. 


科資中心編號: RN9102-0100 
題名: 以實體法為基礎之獨奏型音樂合成器與其對應之硬體架構設計與製作 
研究者: 蘇文鈺 ; 陳永源 
機構名稱: 成功大學資訊工程系 
經費來源: 行政院國家科學委員會 
補助編號: NSC89-2218-E006-131 
研究日期: 2000/08 - 2001/07 
出版日期: 2001/10 
原件提供: 行政院國家科學委員會科學技術資料中心, NSC89-2218-E006-131 
科資分類號: IA0404 
關鍵詞: 實體模型法 ;崁入式滑音架構 ;電子音樂合成 ;硬體描述語言 ;類神經網路 
Physical Modeling ;Embedded Portamento ;Electronic Music Synthesis ;Hardware Description Language ;Neural Network 
中文摘要: 實體模型法是近年來最受重視的新一代的 電腦音樂 合成方式,其擬真度為其他傳統的方法所未及,因此以之來製作獨奏型樂器較其他方法更為適合,可是由於其計算量較高,因此也比其他方法更需要使用專用的硬體來進行合成的工作,在本次的計劃中,我們提出一個以實體合成法為基礎的崁入式的滑音架構,並設計其硬體架構且以硬體描述語言製作之,其結果為一適合多通道之合成器,而且面積相當小而速度快,當與一個小型的微處理機結合後,其成果已經接近可以商品化的標準。 
英文摘要: Physical modeling method is the most important computer music synthesis technique in the past few years. Because it can synthesize very realistic musical tones, using it to construct a solo electronic musical instrument is a better choice compared to other conventional methods. However, due to the higher computational requirement, it requires a dedicated hardware to perform the synthesis works. In this project, we propose an model-based embedded portamento synthesis structure and design the corresponding hardware architecture. The implementation is done by using Verilog HDL and what we design is a multi-channel synthesizer that is capable of synthesizing multiple tones simutaneously.It is compact and high speed. 

科資中心編號: RN9101-0220 
題名: 光學音樂辨識雛型系統之研發 
The Development of an Optical Music Recognition Prototype System 
研究者: 蘇木春 
機構名稱: 中央大學資訊工程系 
經費來源: 行政院國家科學委員會 
補助編號: NSC89-2218-E008-034 
研究日期: 2000/08 - 2001/07 
出版日期: 2001/07 
頁冊數: 4頁 
原件提供: 行政院國家科學委員會科學技術資料中心, NSC89-2218-E008-034 
科資分類號: IA0403;IA0406 
關鍵詞: 光學音樂辨識 ;影像處理 ;點字樂譜 ;圖形識別 ;輔具 
Optical Music Recognition ;Image Processing ;Braille Music Score ;Pattern Recognition ;Assistive Device 
中文摘要: 我們知道盲人若要從事音樂上的工作或興趣的培養,可以說是困難重重。因此,一種可行的方式是替盲人發展出光學式盲用點字樂譜轉譯程式。使用者可以將所要學習的樂譜,先經掃瞄器掃瞄至電腦檔案中,接著藉由這個轉譯程式直接轉譯成點字樂譜檔案,然後,便可以用盲用點字機顯示或製成點字樂譜,如此一來,便可以大量地減少人工製作的費用以及縮短製作時間。而光學式盲用點字樂譜轉譯程式的關鍵是在於光學音樂辨識系統(Optical music recognition ,簡稱 OMR)能否有效運作,惟有能夠正確辨識樂譜,才能將點字樂譜的製成自動化。因此,本計劃希望在一年的期間內,發展出光學音樂辨識雛型系統,這個雛型系統的首要目標是能夠快速且有效地辨識大部份的音樂符號。 
英文摘要: We know that it is very difficult for blind persons to either work as musicians or play music instruments just for fun. One possible solution to the problem is to develop an optical Braille music score translating software. A user can first scan music scores and then run the translating software. The recognition results can be either output to Braille or used to generate Braille music scores. By using the software, we can greatly lower the labor fee and shorten the manufacturing time. The success of the optical Braille music score translating software greatly depends on the performance of the"optical music recognition" systems. Automation of generating Braille music scores becomes possible only if the "optical music recognition" system can efficiently recognize music scores. Therefore, we decide to propose a one-year project to develop an "optical music recognition" prototype system. The main goal of this project is to efficiently recognize most of music symbols in music scores. 

科資中心編號: RB9009-0093 
題名: 虛擬室內環境三度空間音響之研究與實現 
3D Sound System for Virtual Indoor Environments 
研究者: 鄭士康 
Jeng, Shyh-Kang 
機構名稱: 台灣大學電信工程研究所 
經費來源: 行政院國家科學委員會 
補助金額: 新台幣 365 仟元 
補助編號: NSC89-2213-E002-027 
研究日期: 1999/08 - 2000/07 
出版日期: 2000/08 
頁冊數: 5頁 
原件提供: 行政院國家科學委員會科學技術資料中心, NSC89-2213-E002-027 
科資分類號: IE0501 
關鍵詞: 虛擬實境 ;三度空間音響 ;空內聲波 ;空間雙耳轉移函數 ;電波傳播 
Virtual Reality ;Three-Dimensional Sound ;Indoor Acoustic ;Binaural Room Transfer Function ;Wave Propagation 
中文摘要: 隨著電腦多媒體時代的來臨,「虛擬實境」(virtual reality, VR)是目前頗為熱門之研究課題。在聲覺效果的研究發展上,3D音響(3D sound)系統乃是目前電腦音樂之新潮流。本研究實作完成「虛擬室內三度空間聲場」( virtual indoor 3D acoustic environment)系統,為簡易3D音響系統加上殘響、迴音等效果;這些效果可由「空間轉移函數」(room transfer function, RTF)或「空間雙耳轉移函數」(binaural room transfer function, BRTF )表示。本研究中,將提出二個架構:一為利用模擬的BRTF配合DSP技巧,一為利用模擬的RTF配合量測的「頭部有關轉移函數」(head-related transfer function, HRTF)及DSP技巧來實作「虛擬室內三度空間聲場」。 
英文摘要: With the coming of computer multimedia era, 'virtual reality' (VR) has become a more and more popular topic. This research has implemented a'virtual indoor 3D acoustic environment', which combines a basic 3D sound system with reverberation and other effects in order to simulate natural acoustic scenes. These effects can be described by 'room transfer function' (RTF) or 'binaural room transfer function' (BRTF). This research proposes two structures to implement 'virtual indoor 3D acoustic environment system.' One is to combine the BRTF simulation and DSP techniques. The other is to use the RTF simulation, the HRTF measurement and DSP techniques. 


科資中心編號: RE8910-0133 
題名: 音樂合成法之研究 
A Study on Music Synthesizer 
研究者: 洪西進 
Horng, Shi-Jinn 
機構名稱: 台灣科技大學電機工程系(NTSTEET) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 257 仟元 
補助編號: NSC88-2213-E011-082 
研究日期: 1998/08 - 1999/07 
原件提供: 行政院國家科學委員會科學技術資料中心 NSC88-2213-E011-082 
科資分類號: PG0102;IE0306 
關鍵詞: 電腦音樂 ;音樂合成 ;合成音樂 
Computer Music ;Music Synthesizer ;Synthesizer Music 
中文摘要: 由於 電腦音樂 的蓬勃發展再加上電腦科技的突飛猛進和CPU的計算能力大幅的提升,音樂的合成已可以用軟體來取代。人們已不需要利用音樂合成晶片的協助,就可在電腦上為所欲為玩音樂。問題是以前的商業產品其用軟體合成出來的音樂品質並不是很好,所以本計劃主要是對電腦合成音樂的合成方法和音色檔的製作做最深入的研究。最後整合我們的研究結果並在MS-Windows上實作一套符合GM (General MIMD)規格且可以展示的音樂合成系統的雛型。根據前人的研究,我們認識到Wavetable的合成方法可以得到較清析的樂音,故本計畫也朝著這個方向研究。去年我們已經在鋼琴樂器上取下了六個八度音色的Wavetable然後以內差合成的方式合成一首完整的樂曲,就整體聆聽的感覺相當的不錯,但還是達不到盡善盡美的地步。所以本計劃決定研究各種其他改善合成的方法,期以合成出來的音色能達到盡善盡美的境界。 
英文摘要: Owing to the rapidly growing VLSI techniques and increasing the computation power of the CPU, the music synthesizer can be implemented by the software technique Instead of using a VLSI chip. Previously, the music synthesizer implemented by software technique is no good; in this proposal, first we will focus on how to develop various efficient music synthesizer methods and tone file creation. Then a prototype of the music synthesizer will be implemented on the PC under MS-Windows system. According to the results proposed by other researchers, we found that the music synthesizer based on the wavetable method can produce better tone color. Therefore, we will develop this project based on the wavetable method. During the last year, based on the wavetable method we sample six octaves tone sample for the piano. Then using the interpolation method, we synthesize a song. The tone color of the synthesized tone is quite good but not the best. In this proposal, we will improve it and get the best synthesized tone color in the near future. 

科資中心編號: RB8910-0197 
題名: 音樂資料庫系統---內涵查詢、資料探勘及智慧型代理人 
Music Database Systems---Content-Based Querying, Data Mining and Intelligent Agents 
研究者: 陳良弼 
Chen, Arbee L.P. 
機構名稱: 清華大學資訊工程系(NTHECOS) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 746 仟元 
補助編號: NSC88-2213-E007-052 
研究日期: 1998/08 - 1999/07 
原件提供: 行政院國家科學委員會科學技術資料中心 NSC88-2213-E007-052 
科資分類號: IA0304;IA0407 
關鍵詞: 音樂資料庫系統 ;資料探勘 ;智慧型代理人 ;查詢處理演算法 ;資料檢索 
Music Database System ;Data Mining ;Intelligent Agent ;Query Processing Algorithm ;Data Retrieval 
中文摘要: 在本計畫中,針對多媒體資料庫中的音樂資料方面的相關問題,進行深入的研究與探討,具體的研究成果,包括:設計音樂主題的擷取演算法,能夠有效率地由音樂物件的旋律中擷取出重覆出現的主題,發展音樂查詢處理演算法,進行相似性比對,並且建立可以反映音樂樂理的相似性衡量函數及音樂資料庫的實作。主要的研究成果,均已發表在國際知名的期刊與會議記錄上。 
英文摘要: In this project, we focus on research issues of music database system and propose our solutions. The results consist of three parts. First, we devise an efficient algorithm to extract themes from music data. Then, a query processing algorithm is also proposed, as well as a similarity measurement function taking music properties into considerations. We also implement a music database system prototype. Our research work have been published on the proceedings of international conferences. 

科資中心編號: RB8902-0050 
題名: 以遞迴散布式類神經網路進行木管樂器實體合成模型之研究 
Model-Based Synthesis of Woodwind Instruments by Using Scattering Recurrent Networks 
研究者: 蘇文鈺 
Su, Alvin Wem-Yu 
機構名稱: 中華大學資訊工程系(CHUCOS) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 460 仟元 
補助編號: NSC88-2213-E216-015 
研究日期: 1998/08 - 1999/07 
頁冊數: 5頁 
原件提供: 行政院國家科學委員會科學技術資料中心 NSC88-2213-E216-015 
科資分類號: IA0404;IA0407 
關鍵詞: 音樂合成 ;線性散布式遞迴網路 ;虛擬樂器 ;多媒體 ;數位波導濾波器 
Music Synthesis ;Linear Scattering Recurrent Network ;Virtual Instrument ;Multimedia ;Digital Waveguide Filter 
中文摘要: 樂器實體模型分析法所面臨的主要有兩個問題:一、樂器發聲體本身振盪的行為模式;二、實體合成模型的建構。為了能有效模擬木管樂器之實體,依單簧片振動的行為模式,吾人提出一三層類神經網路架構,我們根據數位波導濾波器發展出一套新的類神經網路稱之為Scattering Recurrent Network(SRN),作為木管樂器的實體合成模型。我們採用Backpropagation Through Time(BPTT)與Simulated Annealing Resiliant Back-propagation(SARPORP)之綜合演算法加速類神經網路參數的學習與其準確度。 
英文摘要: There are two primary problems when model-based music synthesis methods are used. First, analysis of vibration of the sounding mechanism of the target acoustic instruments. Second, design of the physical synthesis model for the instrument. In order to synthesizing woodwind instruments, we propose a 3-layer neural network to model reeds used in woodwind instruments and a Scattering Recurrent Network to model the bores. In the training of the synthesis model, a hybrid method which combines Back-propagation Through Time (BPTT) as well as Simulated Annealing Resilient Back-propagation ( SAPROP) is used to achieve fast convergence and accuracy. 

科資中心編號: RE8806-0068 
題名: 大專院校音樂科系 電腦音樂 與電腦輔助音樂教學實施現況之探討 
A Survey on Computer Music and Computer-Assisted Music Instruction Courses at the College Level 
研究者: 李友文 
Lee, Yu-Wen 
機構名稱: 高雄師範大學音樂系(NKNUMUS) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 206 仟元 
補助編號: NSC86-2417-H017-002 
研究日期: 1996/08 - 1997/07 
原件提供: 行政院國家科學委員會科學技術資料中心 NSC86-2417-H017-002 
科資分類號: PG0100 
關鍵詞: 電腦音樂 ;音樂教育 ;電腦輔助音樂教學 ;大學校院 
Computer Music ;Music Education ;Computer-Assisted Music Instruction ;College And University 
中文摘要: 本研究主要針對台灣大專院校音樂科系 電腦音樂 與電腦輔助音樂教學相關課程實施現況作一探討研究。為使研究達到詳細確實,採用到校訪視與訪問(面談或電話)調查法。由研究者親至各校音樂系登錄硬體規格,拍攝幻燈片存證,並向各校擔任該課程之教師們訪問其實際教學情形。所得資料經整理分析後依以下八大重點逐一探討: (1)硬體設備(2)教學目標(3)課程內容(4)教學成效(5)師資(6)學生(7)教學困難(8)發展方向在本次調查全國十七所大專院校音樂系中,截至八十六學年度止,共有十四所設有 電腦音樂 或電腦輔助音樂教學相關課程。一般大學音樂系與師院系統的音樂教育系,由於對學生未來生涯發展方向明顯規劃不同,因此音樂系的課程內容設計傾向 電腦音樂 教學應用,如錄音、編曲、製譜,而音樂教育系所開設之課程較著重如何運用電腦來輔助音樂教學,如音樂教學軟體介紹,評估與應用。後續相關的研究建議如下: (1)研究所階段 電腦音樂 課程之研究。(2)中小學推行 電腦音樂 教學之研究。(3)中小學音樂師資在職進修電腦輔助音樂教學課程之研究。 
英文摘要: The purpose of this study is to investigate the current status of courses in computer music and computer-assisted music instruction that are offered by college music departments in Taiwan. In order to obtain complete first-hand information, the investigation consists of two major parts: the visiting and inspection of music departments and their facilities; the interviewing (either in person or by phone) of the instructors of pertinent courses within the departments. For collecting reference material, during departmental visits, the researcher recorded the specifications of each piece of equipment a given music department has and photographed the equipment itself. Audio recordings of each of the instructor interviews were also made. These interviews include discussions about the details of teaching computer music and computer-assisted music instruction courses. After this data was rigorously analyzed and synthesized, the results were categorized and discussed as follows: (1) equipment (2) instruction objectives (3) course content(4) instructional achievement (5) background of the instructors (6) students' reaction (7) instructional difficulties (8) possible direction of future development Of the seventeen college music departments included in this survey by the end of school-year of 86, fourteen offer either computer music courses or courses in computer-assisted music instruction . Apparently because of differences in the future career requirements of students, the content of courses offered by music programs and music education programs are different. Music programs tend to provide courses that deal more with sampling, sequencing, recording, arranging, and notation etc., whereas music education programs offer courses which emphasize computer-assisted music instruction. Suggestions for future studies are listed as follows: (1) computer music curriculum at the graduated level (2) computer-assisted music instruction at the elementary and secondary school levels (3) courses in computer-assisted music instruction specifically designed for the continuing training of elementary and secondary school music teachers. 

科資中心編號: RE8701-0115 
題名: 音樂資料電腦化系統之研究 
An Exploratory Study of Computerized System in the Music Library. 
研究者: 陳郁秀 ; 賴麗君 
Chen, Yuh-Shiow ; Lay, Lih-Jiun 
機構名稱: 台灣師範大學音樂研究所(NTNUMUSG) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 490 仟元 
補助編號: NSC84-2413-H003-032 
研究日期: 1995/06 - 1996/07 
頁冊數: 77頁 
原件提供: 行政院國家科學委員會科學技術資料中心 NSC84-2413-H003-032 
科資分類號: PG0100 
關鍵詞: 音樂圖書館 ;線上公用目錄 
Music Library ;Online Public Access Catalog (Opac) 
中文摘要: 本研究計畫是以音樂教育普及化為終極目標。對象不局限於音樂專業人士,也包含業餘音樂愛好者。由於師大音樂系圖書館典藏豐富,並具備音樂專業人士從中輔導,因此將以師大音樂系圖書館為雛形,藉以發展應用至國內各其他圖書館。目前國內音樂科系和藝術學院相繼成立,年輕的音樂家、學成歸國人數也日益增加,無形中帶動了音樂風氣的蓬勃發展。而國內音樂家的素養已達國際水準,但於音樂視聽及圖書方面的設備卻被忽視。現歐美各國皆有其代表之圖書館,如美國紐約巿立圖書館、大英圖書館等都是提供音樂學術資料的極佳來源,對提昇音樂知識有莫大的助益。國內目前犯罪率節節上升,而音樂的提昇可陶冶性情,與潛移默化之作用。在此情況之下,一個完整的開架式音樂圖書館實是刻不容緩。例如音樂家華格納為了讓觀眾瞭解他的歌劇而著作「音樂與戲劇」等書。而音樂圖書館的必要不僅為音樂科系,而且要擴及社會民眾,以豐富他們的音樂知識。國內除各大專院校所屬之圖書館外,各省巿也有所屬之圖書館,然而由於涵蓋範圍過於廣泛,各系專業資料難免不盡完備,常造成使用者之不變。在系統編纂方面,鑑於歐美各國採用「國會圖書分類編目系統」以取代傳統式「杜威系統」。因此本計畫將依據此基礎研究出一套便利的電腦化資料查詢系統,並作為其他各圖書館之典範。在現有資料中我們將作系統的整理並作探索性的研究,且與國內外各圖書館作比較與評估,並考量讀者之需求,透過問卷實地調查和座談方式共同研討改進之方針。有了完整而系統化的音樂圖書館,不僅嘉惠音樂學界及一般民眾使用,對提昇學術研究大有助益;同時透過網路連線,也可以使國內外人士的專業知識傳遞更加迅速。此外,國外學者從我們圖書館裡的民族音樂館藏,也可以對中國音樂有更深一層的認識。 
英文摘要: The goal of this project is an education purpose for not only professionals but also amateurs. The music activities in Taiwan now are growing enormously which attract both musicians and music lovers. The well-systematic music library is urgently demanded. The music libraries in all the universities or conservatories are not organized professionally owing to the limit of professional knowledge and some technical problems. Under this circumstance a high standard filing system has to be formed. The tasks of this project toward several aspects regarding contents, facilities and building a catalogue system. In the aspect of contents include rare collections, manuscripts especially the Ethno-Chinese Music. To build a catalogue system the library will be computerized, which means that cataloging, circulation etc. will all run through the computers. The main body of this project concentrates on setting up a filing rule in order to build up a professional standard in cataloguing. As far as the filing system is concerned, the traditional Dewey Decimal Classification has been replaced by the Library Congress System in most countries. Therefore we are going to base on this ground to develop a convenient classification for our music library, which serves as a model for the other music libraries. For the methology this project will start with the arrangements of the materials. The first step will begin with our own library, then compare and evaluate with its system at the other universities or conservatories. Secondly, an opinion poll will be progressed on the readers, in order to see their demanding and difficulties. Moreover a survey should be mailed to the readers, in order to analyze any defects of the system. Survey questions focused on the topics such as the age of the readers, the social standard, or the frequence of using the libraries etc... Owing to the differences of the readers' requirements between Taiwan and foreign countries, symposiums with professionals and specialists should be organized for exploratory study and review. The idea of this computerized system intends to abridge the gap between the readers and the equipments. 

科資中心編號: RE8702-0091 
題名: "單音結構"的美學基礎與其應用模式的建立
The Aesthetical Foundation of a Single Tone and the Construction of Its Application Model. 
研究者: 吳丁連 
Wu, Ting-Lien 
機構名稱: 交通大學應用藝術研究所(NCTUARTG) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 348 仟元 
補助編號: NSC84-2413-H009-003 
研究日期: 1994/08 - 1995/07 
原件提供: 行政院國家科學委員會科學技術資料中心 NSC84-2413-H009-003 
科資分類號: PG0100;PG0002 
關鍵詞: 單音 ;心理音響學 ;機制 ;一元論 
Single Tone ;Psychoacoustics ;Mechanism ;Monism 
中文摘要: 本專題是從不同的學術領域---物理音響學、心理學音響學、文化現象、美學、創作理論等方面,去探討單音結構,進而建立二種帶有理念性質的單音應用模式。整個專題報告主要內容包括: 1.音與心之機制:從音的面面觀(物理的、心理的、文化的、審美的)、心的作用(知覺、認知、觀察、觀照),最後落在心物合一之一元論點---音即心、心即音。2.審音以知樂:以音與心之機制作為背景,進而闡釋兩種美學範疇:一即多,多即一;同與異(統一與變化),說明審美客體自身的不存在性。知覺上,一個"有意味的"整體性,來自兩方面的作用:一方面為"幻相客體"。它是由內在成分之異同所形成的關係網絡而得;另一方面的作用,則來自主體對文化文本的知覺與認知。3.應用模式之建立:以單音在時間流程裡的輪廓:頭、(體)、尾,和單音的走音方式:落-行-收,作為單音模式。分別依一般樂器環境,和 電腦音樂 環境,建立二種不同的應用模式,以為音樂創作和闡釋之理論基礎。4.附錄:(1)此專題研究的實驗 電腦音樂 環境---Kyma之介紹,和部分實驗成果之說明(以英文書寫)。(2)此專題研究專門術語之中、英文對照。(3)相關書目。(4)古琴音樂單音的八十八種音色。整個研究圍繞在「審音以知樂」,和「破微塵而出大千經卷」的前提下,企圖建構或解讀「一音一宇宙,一曲一乾坤」的命題。 
英文摘要: The bottom-line of this monograph is to establish ideal applied models for "single tone" based on the research done here of its structure in different fields of study, such as the physical acoustics, psycho-acoustics, cultural phenomena, aesthetics, and compositional theories. The primary contents of the monograph are: 1. The mechanism between the "tone" and the "mind". From the observation of the "tone" in various aspects (with the physical, psychological, cultural, aesthetic point of views,) to the functions of the mind (perception, recognition, observation, and all-awareness) comes the view of monism for the fusion of the tone and the mind, which asserts the tone being the mind, and vice versa. 2. Knowing music by investigating the tone. Two categories (one as many, many as one and unity/variation) in aesthetics are delineated with the mechanism between the tone and the mind as the background. A perceived significant completeness is the outcome of the following: In one way, it's the perception and recognition followed when the subject is confronted with the cultural context. 3. The establishment of the applied models. The models for"single tone" is structured after the formation process of a single tone in the time flow, referring to the head, body, tail outline of a tone from its beginning to decay. Two types of models are established as for one's compositional as well as theoretical basis: one for the environment of traditional instruments, and the other for that of computer. 4. Appendix: (1) Introduction of "Kyma system", the experimental computer environment being used in this research. (2) terminology (Chinese/English comparison) (3) Bibliography (4) different timbres of a single tone on the Chinese instrument---"Chin" The entire research aims to establish or interpret the thesis of "a single tone as the cosmos". 

科資中心編號: RC8607-1068 
題名: 視窗環境下的音類理論/作曲環境
A Computer Environment for Pitch Class Theory and Composition. 
研究者: 吳丁連 
Wu,Ting-Lien 
機構名稱: 交通大學應用藝術研究所(NCTUARTG) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 321 仟元 
補助編號: NSC83-0301-H009-024 
研究日期: 1994/02 - 1995/01 
原件提供: 行政院國家科學委員會科學技術資料中心 NSC83-0301-H009-024 
科資分類號: PG0101 
關鍵詞: 音類集 ;集類 ;組合律 ;衍生音列 
Pitch-Class Set ;Set Class ;Combinatoriality ;Derived Series 
中文摘要: 此研究一"視窗下的音類理論/作曲環境"是以80年國科會專題計畫---"電腦上的音類集環境"為基礎,進一步加以更改與擴充而成。(一)電腦部分:1.環境架構---從原先封閉式的DOS環境改為開放式的視窗環境,其理由為:原先DOS環境的封閉性特色不易作為輔助教學和電腦教學環境所用。而視窗環境的整合性與個體導向之特色,使它易與其他音樂軟/硬體(如合成器、音效卡、記譜軟體)相結合,而形成靈活使用且完整的環境。2.程式內容---由原先著重音類結構的無序性面(音類集)之運算,擴充到它有序性面(音類節、音類環、音類系列)的處理;也從原先僅講究音類理論實用運算的環境,擴及到可被聽得到的音類結構之作曲環境,和可一步步介紹音類理論的輔助教學環境。(二)研究內容:由原先著重音類無序性面的探討,擴及到音類有序性面的研究,且進一步探索作曲前可能性的音類結構設計與意識。整個研究的內容,概述如下: 1.音之基本:音在此研究是指可感知的音,為了理論上的探討,將它的特殊性面相---含有音域空間的稱之為音高,而它的一般性面相---帶有抽象思維性質的稱之為音類。所以音類可說是音的種類,但有時也可說是音高的種類,兩者意義一樣,只不過有時因兩者處的前後文不同而有所不同。同樣的,一群音稱之為音群,從它的特殊性面相看,稱之為音高群,而從它的一般性面向來看,稱之音類群。在"音之基本"的探討裡,包括:(1)音、音高、音類、有/無序音高程、有/無序音類程的認識與它們的數字與符號之表示;(2)音群、音高群、音類群它們的種類、定義、表示方式及音程內容。音高群與音類群一樣,皆可從群之份子有序/無序的角度、去進一步界定它們各自的種類。有序音高群稱之音高節、音高環、或音高系列,無序音高群稱之音高集;有序音類群稱之音類節、音類環或音類系列,無序音類群稱之為音類集。"節"表示音份子線性的次序性,有起點與終點。"環"表示音份子具有循環的次序性,無起點與終點。"系列"與"節"一樣,表示份子線性的次序性,不同的是它的使用常與系列音樂聯想在一起。"集"表示音份子之無序性,此概念來自數學集合論,其前提為:以相同的音份子在任何次序下組成的不同集合形式,皆具相同意義。(3)音高群、音類群之基本運算與它們的表示符號。2.無序音類群(音類集):(1)音類集的代表形式與記法內容,包括音類集的遞增形式、正規形式、移位代表形式、轉位代表形式和移/轉位代表形式;(2)音類集和它移/轉位形式之共同音關係;(3)音類集之間的含攝關係---包括宇音類集、子音類集、母音類集、巾音類集、聯集、補集、交集、移位共同音向量,轉位共同音向量;(4)共同集類間的關係:(a)"Z"關係、"M"關係、"MZ"關係與它的類似性關係,音類份子最大類似性,R/sub p/和音程類向量的類似性(R/sub 0/、 R/sub 1/、R/sub 2/); (b)共同子集數(TMEMB)與共同子集值(ATMEMB)、比較共同音向量(CMNV)和同音數目向量;(c)複集(set-complex)與次複集(set- subcomplex)。3.有序音類群(音類系列):(1)有序音類群、系列、十二音系列、十二音音樂、十二音聚團的基本認識,包括十二音系列之背景,基本假設及其在音樂的結構角色;(2)十二音列數字記譜的種種,包括音類份子、音類次序、系列形式之數字標示;(3)十二音列的基本運算與符號表示,包含移位、轉位、逆行,及其12*12的矩陣和系列形式之符號標示;(4)一些有特性的十二音列之設計,包括均衡性、所有音程、均衡---所有音程、衍生性、組合律等特性之系列; (5)十二音音樂"系列"思維的擴充,包括"短"和"長"之系列,複合系列,輪轉形式、附屬系列形式,倍數(7)系列形式;(6)系列的共同成分,包括單一系列裡的共同成分,系列形式間的共同成分,同一轉位指數的系列形式和不變的分割次序。4.作曲---15種基本音類結構之設計與意識,說明作曲前音類意識的可能性活動。每個設計可獨存於任何可能大小的音樂意義單元,也可與其他設計相互滲透、合併、組合於任何可能性的意義單元。5.電腦上的音類理論/作曲環境,包括: (1)電腦輔助教學環境---將此研究的音類理論概念,透過電腦的環境來完成教學活動。(2)作曲家工具箱,以簡單圖表顯示作曲家在作曲前對音類結構之可能性資料,如系列的12*12矩陣之48種基本形式,M系列形式,R系列形式.....等。(3)作曲環境---提供音類結構的設計,配合音樂其他變數的活動,透過 電腦音樂 環境,呈現可聽得到的聲音結構,進一步透過此環境所提供的語法,發展、變化、組織成一首曲子。曲子的記錄可透過與此環境相連接的記譜軟體或錄音設施來作不同的保留。最後是附錄,它提供了本研究之專有名詞中英文對照表、英中文對照表、符號說明、音類理論基本概念的流程表、相關書目和此研究的電腦程式列表。音類理論/作曲在國外(尤其是美國地區)已有60年以上的發展、研究,然而有關這方面研究在國內才開始。此研究為國內首次有關這方面的研究,也是國際上少數應用電腦環境來從事這方面研究者之一。有關中文專門術語在缺乏共識下,作者藉助與此研究相關的跨科領域,數學集合理論和概念邏輯的中文學術術語作為參考之依據。 
英文摘要: possible by the notational software connected with this environment. At last is the appendix, which provides the Chinese-English and English-Chinese comparison chart for the terminology used in this research, description of the symbols used, a flowing chart for the main concepts discussed in this research, bibliograghy, and the computer programs of this research. Pitch-class theory/composition has been developed and studied for over 60 years abroad (especially in U.S.A.), though it has just started here in our country. This research is the ever first study in this area in our country, and is also one of few organizations or individuals who devote themselves in this area by the application of computer. For lack of the Chinese terminology with common recognition, the writer turns to the terminology used in the "set theory" of mathematics and "Aristotelian logic" as his point of reference. 

科資中心編號: RC8501-1041 
題名: 電玩式輔助學習之研究---使用信號處理於高中音樂教學
Game-Oriented Computer-Assisted Learning: Using Signal Processing for High School Music Course. 
研究者: 劉睦雄 
Liu,Mu-Shieung 
機構名稱: 台灣大學資訊工程系(NTUECSIE) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 365 仟元 
補助編號: NSC84-2511-S002-008-CL 
研究日期: 1994/08 - 1995/07 
出版日期: 1995/07 
原件提供: 行政院國家科學委員會科學技術資料中心 NSC84-2511-S002-008-CL 
科資分類號: SC0103;IA0408;PG0100 
關鍵詞: 數位信號處理 ;電腦輔助教學 ;音樂教育 ;電腦音樂 ;高級中學 
Digital Signal Processing ;Computer-Assisted Instruction (Cai) ;Music Education ;Computer Music ;Senior High School 
中文摘要: 過去的電腦輔助音樂教學一直受限於工具的採用,人機介面多使用鍵盤、滑鼠等設備。近年來由於個人電腦上類比轉數位的介面已成熟,使得用人聲做為輸入介面成為一種可能。本報告主要目是研究如何將樂音的音高(pitch)特徵抽取出來,以做為後續處理的基礎。包括自我相關函數(autocorrelation)、修正窄自我相關函數(narrowed autocorrelation)、傅立葉轉換(Fourier transformation)和constant-Q transformation都將被探討。另提出一演算法,做為多音抽取的基礎方式。音高抽取除了在教學方面有所貢獻外,其抽取出的資訊可用來做為合聲分析、曲式分析、自動謄譜系統及輔助作曲等應用。對於發展 電腦音樂 的研究而言是一項基本的工具。 
英文摘要: In the past, keyboard and mouse are the main tool for user interface of the music education. Nowadays, because of the cost down of the AD interface of the personal computer, it is possible to use voice for input interface. The object of this report is to discuss several methods about how to extract music pitch from the analog music signal. The topics include autocorrelation function, narrowed autocorrelation function, Fourier transformation and constant-Q transformation. An algorithm for extracting multi-pitch is provided for reference. Pitch extraction is not only applied on music education but also useful in harmonics analysis, style analysis, auto- transcription system and computer-aided composition. This is a basic tool for research in computer music. 

科資中心編號: RC8410-1040 
題名: 整合電腦作曲環境之研究 
The Research of an Integrated Environment for Computer Music Composition. 
研究者: 吳丁連 
Wu,Ting-Lien 
機構名稱: 交通大學應用藝術研究所(NCTUARTG) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 558 仟元 
補助編號: NSC80-0301-H009-04 
研究日期: 1990/08 - 1991/07 
出版日期: 1994/04 
科資分類號: PG0101;IA0411 
關鍵詞: 準則作曲 ;電腦音樂 ;聲音個體 
Algorithmic Composition ;Computer Music ;Sound Entity 
中文摘要: 此研究報告包括三個部分:(一)文字部分涵蓋(1)中/英文摘要;(2) 電腦音樂 的回顧與探討;(3)準則作曲整合環境的系統說明;(4)整合作曲環境之使用手冊;(5)在此整合作曲環境之創作實例與說明;(6)附錄,包括參考書目與整合作曲環境之原始程式碼。(二)軟體,在NeXT電腦上建構一整合電腦作曲環境。(三)譜例與錄音帶。此研究的目的在以結合交大 電腦音樂 工作室現有的硬體設備及發展中的作曲與音類集理論準則,到達一個整合的,及時的,圖形化的,能剪輯實驗的,交互的作曲環境。在此整合作曲環境中,對素材處理的態度是視一個單一的聲音、樂逗、樂句、樂段或樂章,為大小不同、各自獨立的單一聲音客體。也基於此,整個環境可依使用者的需要分為兩大類---素材與語法:(一)素材方面:(1)使用者可依自己的想法從電腦鍵盤或合成器鍵入基本動機音形(兩個音或兩個音以上),藉著此環境提供的準則或自設準則來產生聲音事件,(2)這聲音事件可被形變成新的音樂材料被儲存為另一聲音事件;(二)語法方面:(1)提供準則,或提供密碼圖形讓使用者設計準則去擴充單一聲音客體到更大範圍的單一聲音客體。(2)提供不同組織單一聲音客體的方式,包括擴充、縮小、並列、重疊、差補、變形...等等。此整合環境建立在NeXT電腦上,結合NeXT現有的軟體,提供DSP(DSP5601 Chip)或外接合成器產生音源。一位使用者可以在此環境很快地藉著音形體現他音樂實驗性的想法。而在音形變化的過程中,或音/音群間的彼此層層相疊、前後相隨的連接過程中進而具體化使用者的想法成一首作品,或藉此擴充音樂經驗空間。整個環境的架構,主要是來自物件導向的觀念,以開放結構為主。所以此環境的語法基本上是孤立性的語法或說蒙太奇,或拼貼性的手法。而不是以一種因果關係,單一焦點發展的封閉系統之邏輯語法。然而此環境並不排斥一位使用者邏輯因果語法的介入。這個環境可能是國內首次對電腦作曲環境的研究與建構,它可帶給台灣音樂/創作/教學的一個具有潛力的實驗性環境。 
英文摘要: This research project comprises three parts: I. The Research Report which includes (a) Abstracts in both Chinese and English, (b) A review and discussion on Computer Music, (c) A description of the Integrated Computer Music Environment(IACE)'s System Architecture, (d) User's Manual of IACE, (e) Some musical examples with the IACE environment, (f) Appendixes which include references and the software's source codes. II. The Software: The Integrated Computer Music Software Environment built on the NeXT Computer. III. Scores and Tapes. The project is researched and developed under the Computer Music Studio of National Chiao Tung University. The motivation of this research project is to combine the studio's hardwares with the composition theories of pitch-class set theory and algorithmic composition developed in earlier projects. The goal of this project is to develop an integrated, real-time, interactive and user- friendly experimental composing environment supporting various kinds of editing functions. In the IACE environment, musical materials such as a single tone, or musical figures, sentences or sections with different sizes are treated as independent sound entities. Based on this concept, there are two major components in this system as explained below: I. Materials: a) User gives a motivic figuration (e.g. 2 or 3 notes) from the computer keyboard or any MIDI devices, and hence generates sound events from the system's algorithms or any user- defined algorithms. b) User can then manipulates these events with all sorts of editing functions, also storing or retrieving them. II. Syntax: a) System provides algorithms, and graphical interface for defining customized algorithms in order to expand a single sound entity into a larger context. b) Provide various functions for organizing sound entities, such as expansion, shrinkage, superposition, juxtaposition, complementation, and other transformations etc. The whole software system was built on the NeXT computer, taking advantage of the computer's flexible interfaces and its DSP56001 chip for direct sound processing. In this environment, users' musical imagination or exploration can be crystallized quickly, furthermore, users is able to capture the structural relationship and coordination between sound entities, easily organize them into a piece of composition. Under the environment, user benefits from expanding his musical experiences. The architecture of the environment is based upon the concepts of Object-Orientation. Therefore, the environment is an open-system that supports montages or collages of isolated syntax, in contrast with traditional single-focused, causal-relational close-system syntax and logic. Of course, under the environment, the user is free to construct syntax of any kind. The software environment constructed for research purposes in the field of computer music is one of the first in Taiwan. Exploring into the rich potential of new musical fields, it shall contributes greatly to the development of Taiwan's experimental music. 

科資中心編號: RC81-1189 
題名: 電腦輔助音感教學軟體之設計與發展 
The Design and Development of a Computer-Assisted Instrution in Music; Aural Training. 
研究者: 賴美鈴 
Lai,Mei-Ling 
機構名稱: 台北師範學院音樂教育系(NTTCMUE) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 252 仟元 
補助編號: NSC80-0301-H152-03 
研究日期: 1991/02 - 1992/01 
出版日期: 1992/01 
科資分類號: SC 
關鍵詞: 電腦輔助教學 ;音感教學 ;音樂教學 ;課程軟體製作 
Cai ;Aural Training ;Music Teaching ;Courseware Design 
中文摘要: 本研究為發展一套適用於國小三年級的電 腦輔助音樂教學音感課程軟體,目的包括下列 各項:(一)探索發展電腦輔助教學課程軟體的方 法與步驟; (二)分析和設計音感教學之內容; (三)編寫音感教學軟體之腳本;(四)發展音感教學之軟體;(五)探討電腦輔助教學在音樂科應用的可 行性。本研究依據國立編譯館主編之國小音樂課 本第一、二冊,採用電腦輔助教學設計的反覆 練習式為理念,設計並發展音感教學軟體。課 程軟體分第一、二冊,每冊依難易度分成四個 單元和綜合練習五個部分,每部分均包含三個 練習項目:長和短、高和低、填填看。腳本和軟體設計完成後,均經由專家評估 並修訂,最後製作成「音感園地」軟體一套四 片,兩片單色版,兩片彩色版。為了測知本課程 軟體的適用性,研究者於北部和南部各選取國 小三年級學生一班進行測試。由受試者填寫的 使用意見問卷調查表中顯示,受試者對「音感 園地」軟體持正面的態度,且大多數受試者表 示用電腦做音感練習比在教室練習有效,顯示 出電腦輔助教學在音樂科應用的可行性。由於國內尚無類似的軟體,研究者認為「 音感園地」課程軟體僅是一個起步,並建議:繼 續發展各年級的音感教學軟體或其他音樂課程 軟體;增加電腦硬體設備,如:語音卡、滑鼠、耳 機等;舉辦研習會;鼓勵 電腦音樂 教學的研究及 課程軟體的開發。本研究成果除了研究報告外,尚包括磁片 一套四張,腳本兩冊。 
英文摘要: The purpose of this study is to develop an aural training courseware for the third-grade music class. The scope includes: (1) investigating the design and development a CAI courseware; (2) analyzing and designing aural training curriculum; (3) designing the script; (4) developing the aural training courseware; (5) promoting the implementation of CAI in music teaching. The courseware is based on the elementary music text book one and two, published by National Institute for Compilation and Translation. It uses the drill and practice design of CAI. The courseware is devided in two volumes, each volume contains four- lessons and one review. Three types of exercise in each lesson are (1) rhythmic pattern identification; (2) melodic pattern identification; (3) melodic dictation. Both script and courseware have been evaluated by CAI experts and music scholars. The courseware is stored in a set of four diskettes, two for monochrome version, two for color version. For the test of courseware, the researcher selected a group of 87 third- grade students from two elementary schools to work with the computer program. All students were asked to fill out a questionnaire that helped the researcher assess their attitudes about using the computer for aural training drill. The results of the questionnaire was positive. Most students indicated that using CAI in aural training is better than traditional aural training drill. This shows that the use of CAI aural training drill is worth of promoting. Currently, there is no other courseware like this aural training courseware. The researcher hopes to continue this courseware for other grade level and also to design other music courseware. The recommendations of this study contain: (1) to increase the soundsource of the computer class: (2) to offer the workshop in CAI for music teachers; (3) to sponsor the research in CAI in music; (4) to promote the design of music courseware. The results of the study include report, a set of 4 diskettes, and 2 sets of scripts. 

科資中心編號: RC75-6024 
題名: 中小學電腦素養教學研究計畫 
研究者: 王弘宗 
機構名稱: 國立清華大學 
經費來源: 行政院國家科學委員會(NSC) 
補助編號: NSC75-0111-S007-03 
研究日期: 1986/02 - 1987/01 
出版日期: 1987/03 
頁冊數: 59 
關鍵詞: 電腦教學 ;資訊 ;流程圖 
Computer Literacy ;Syllabus ;Pedagogy ;Flowchart ;Loop 
中文摘要: 本實驗的目的在規劃,並研究一套適合新 竹科學工業園區實驗高級中學國小部、國中部 及雙語部之電腦教學課程。長程目標則在歸納 出合理的電腦教學方法,並設計一套適合中、 小學生程度的課程與教材,以因應資訊時代的 變遷。為了不影響正常教學,因此以課外活動的 方式進行實驗研究。在課外活動中成立電腦研 習組,再從有興趣學電腦的學生中,以隨機抽樣 方式選取參加。分成國小及國中兩種程度,國 小程度又分為四、五、六年級三個班次,國中 程度則分為國一及國二兩個班次,此外又增加 雙語部班次,依這些不同程度之班次,實施不同 程度之電腦教學課程,佐以實習輔導,引發學生 對電腦的興趣、認識電腦,並獲得基本的電腦 知識,進而應用電腦解決或協助做一些事情。由初步實驗,發現國小學生較活潑,上課比 較熱烈,又很喜歡問問題,所以老師以問答方式 上課較理想;但是,國中生就不一樣,他們喜歡聽 課,而少發問。從小學四年級到國中二年級都 很難接受流程圖的觀念。普遍的反應都無法接 受迴路(FOR...NEXT)的觀念。學生最感興趣的還是 電腦音樂及繪圖。 
英文摘要: The objective of this study is to develop an appropriate syllabus and pedagogy for computer literacy course to be taught at elementary and middle school, based on the experiment currently carried out at the Science-Based Park School. Our participants, being randomly sampled from among interested students, ranging from 4th grade to 8th grade, are grouped into different classes by their grades, with lecture materials properly adapted accordingly. The course, being on an extracurricular basis, is about half hour lecture followed by another half hour hand-on experiment in the workshop. From this premiere, we find out that lower-division students are relatively more responsive to the question-answer method in the class as compared to the upper-division students. Some notions, such as flowcharting and looping, are generally deemed hard; while computer music and graphics are always found catchy among students. 


科資中心編號: RC8601-1117 
題名: 電腦輔助音樂教學之音源研究 
The Effect of the Qualities on the Computer Assisted Music Multimedia Program. 
研究者: 潘世姬 
Pang,Shyh-Ji 
機構名稱: 藝術學院音樂系(NIAGMUS) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 277 仟元 
補助編號: NSC84-2511-S119-001-CL 
研究日期: 1994/08 - 1995/07 
出版日期: 1996/06 
原件提供: 行政院國家科學委員會科學技術資料中心 NSC84-2511-S119-001-CL 
科資分類號: SC0103;PG0100 
關鍵詞: 電腦輔助教學 ;音樂教學 ;合成音源器 ;音樂軟體 
Computer-Assisted Instruction (Cai) ;Music Instruction ;Synthesizer ;Music Software 
中文摘要: CAL(Computer Assisted Learning)音樂軟體的製作過程中,音源的使用是不可或缺的,而其聲音主要是來自合成音源器(Synthesizer)。音源技術的變革自1967年起歷經FM(Frequency Modulation), PCM(Pulse Code Modulation),取樣方式(Sampling), RCM(Realtime Convolution & Modulation),以及最近的 SWS(Spectral Wavetable Synthesis)等四階段。音源作為人類聽覺感知的主要媒介(S. Mcadams & A. Bregman,1979),其品質對於一個以聽覺訓練為主的教學是最基本而首要的一環。音源選擇的不當將造成學習者在學習上的困擾,進而影響CAL的學習成效。聲音的特質主要分為兩大類:一為物理特性,二則為心理特性。在物理特性方面,聲音是藉由某種物質(如空氣)傳遞而感知的振動現象。基本上,這個振動現象可用三個度量來表示,它們分別為:振動強度、振動速度、振動波形,也就是一般我們所談的:聲強(Intensity)、頻率(Frequency)、以及波形(Waveform)等三種聲音的物理特性。至於聲音的心理特性主要是來自人類的聽覺,可分為以下三點:響度(Loudness)、音高(Pitch)、音色(Timbre)。本實驗的重點主要針對心理特性中的音色部分作一探討,至於其餘兩點:響度和音高則成為實驗中控制不變的因素。此外,本實驗的音源將從合成音源器選取,而如何引介Acoustic聲音的觀念,進一步在合成音色應用轉移,也是此次實驗所嘗試突破的課題。實驗的合成音源來自:Proteus-2(E-mu)、TX802(Yamaha)、TG77(Yamaha)三台機器,這三台音源器分別採用不同的發聲技術---包括取樣(Proteus-2)、FM(TX802)以及RCM(TG77)等三種方式。藉由對音樂系學生(已接受過音樂專業訓練之人員)大量的測試結果,篩選出最適合CAL音樂軟體製作的音色,總合並作出整理,成為日後軟體音源部分製作的資料依據。 
英文摘要: The use of computers are becoming popular in the R.O.C. Applying computers to educational purpose have gradually become a trend. There are much uses of computer in all sectors of society, such as education, home entertainment, business, etc. Although the use of computer assisted instruction (CAI) has gradually reached its maturity, applying computer technology upon musical education appears relatively to have receives relatively little attention. Until now, the CAI programs for music use rate only 5% of the total programs as sponsored by the National Science Committee. Among these music programs, the study of the effect of tone qualities in the computer assisted music multimedia program is frequently ignored. This proposal plans to study the tone qualities in the use of computer assisted music multimedia program by ways of the MIDI method The technological changes in the synthesis of tones have been in the following four phases since 1967: FM, RCM, Sampled sound, and SWS. Tone as a primary hearing sensation of human being, is the most fundamental and indispensable element in computer assisted music multimedia program. The inappropriate use of the tone will deter students from learning, and further affect the learning of CAI programs. This research, which is a novel approach and is new to the ROC, will use MIDI to generate tones through four different tone generators. Therefore, it is hoped that the computer assisted music multimedia programs will benefit from the results of this research.

科資中心編號: RS9010-0016 
題名: 超文件資料庫管理系統之研製(II) 
Design and Implementation of a Hypertext DBMS 
研究者: 劉寶鈞 
Liu, Baw-Jhiune 
機構名稱: 中央大學資訊工程研究所 
經費來源: 行政院國家科學委員會 
補助金額: 新台幣 401 仟元 
補助編號: NSC84-2213-E008-004 
研究日期: 1994/08 - 1995/07 
出版日期: 1995/08 
頁冊數: 66頁 
原件提供: 行政院國家科學委員會科學技術資料中心, NSC84-2213-E008-004 
科資分類號: IA0304 
關鍵詞: 超文件系統 ;資訊存取 ;結構查詢語言 ;查詢圖樣 ;查詢處理 ;資料庫管理系統 
Hypertext System ;Information Access ;Structural Query Language ;Query Pattern ;Query Processing ;Database Management System (Dbms) 
中文摘要: 超文件系統(hypertext systems)為讀者與作者提供了一個便利的環境,幫助他們組織及處理本文資訊。其內部包含一個文件的網路(network)結構,通常此網路結構都是相當巨大且複雜;因此,資訊存取(information accessing )在超文件系統中是一項重要且困難的課題。結構查詢(structural query)為超文件系統中的資訊存取,提供一個有效的方式。本報告提出超文件系統的一個結構查詢語言;此查詢語言可充分且結構分明地表達出超文件系統中結構查詢的各類條件,並且定義一些常用的查詢圖樣規格,使得與之對應的語法足以描述;在查詢處理上,本報告提出兩種查詢處理的策略,其中基本演算法可有效解決查詢條件單純的查詢處理;另一種為遺傳演算法查詢處理策略,可適用查詢條件較複雜的查詢處理,以及一些特殊要求的查詢。從本報告中的實驗結果可知,此兩種查詢處理策略能提供有效的處理方式。 
英文摘要: Hypertext systems provide readers and authors with convenient environments to organize and manipulate text information. The architecture of hypertext systems is networks of text nodes connected by typed links. Usually the networks are large and complex; thus, information accessing in hypertext systems is an important but difficult issue. Structural query suggests an efficient strategy for information accessing in hypertext systems. The report proposes a structural query language for hypertext systems. The query language describes sufficiently and clearly each kind of conditions in structural query, and defines some usual query patterns so that the corresponding syntax can pertinently describe them. For query processing, the report suggests two strategies. One is a basic algorithm to be able to efficient process the queries, of which query conditions are more simple and the other is genetic algorithm strategy, which is suitable to the query with more complex conditions or special requests. The experiment result has shown that the two query processing strategies supports efficient query processing. 

科資中心編號: RB9009-0182 
題名: 寬頻網路電腦協同作業式多媒體虛擬課程之發展環境
A CSCW Multimedia Virtual Course Development Environment on Broadband Networks 
研究者: 施國琛 ; 顏淑惠 ; 王俊嘉 
Shih, Timothy K. ; Yen, Shwu-Huey ; Wang, Chun-Chia 
機構名稱: 淡江大學資訊工程系 
經費來源: 行政院國家科學委員會 
補助金額: 新台幣 925 仟元 
補助編號: NSC89-2213-E032-021 
研究日期: 1999/08 - 2000/07 
出版日期: 2000/09 
頁冊數: 5頁 
原件提供: 行政院國家科學委員會科學技術資料中心, NSC89-2213-E032-021 
科資分類號: IA0408;IA0202 
關鍵詞: 虛擬課程 ;電腦支援協同作業 ;網頁文件 ;寬頻網路 ;多媒體 ;網際網路 ;資料庫管理系統 
Virtual Course ;Computer Supported Collaborative Work (Cscw) ;Web Document ;Broadband Network ;Multimedia ;Internet ;Database Management System (Dbms) 
中文摘要: 在我們的日常生活中網際網路已扮演了一個重要的角色,而且網際網路(Intern et)上所展現的網頁文件已是一個新的通訊與商業上傳播的介質,所以網路教學也因應而生,因此建立一個教學課程的網頁文件變的更加的重要。本論文則提出了一個網頁文件(Web document)軟體發展架構,在與電腦協同作業(CSCW)結合教學課程資料庫的環境下,提供給虛擬大學課程設計者去發展和測試他們所設計的課程單元。而我們設計的課程資料庫的則支援了我們的網頁發文件發展環境(Web document developing paradigm)。 
英文摘要: Web documents are increasingly playing an important role in our daily life, as the Internet has become a new medium for communication and commerce. So internet training become popular now and it is important to build a web document for learning course. In this paper, we propose not only a Web document developing paradigm as well as a CSCW(Computer Supported Collaborative Works) environment with distributed course database support for instructors of a virtual university to develop and test their course materials but also a virtual course DBMS supporting the paradigm. 


[科資中心編號]RB8803-0072
[題    名]
動態異質性資料庫系統中維持資料一致性的交易管理模式的探討
An Investigation of the Transaction Management for Active
Heterogeneous Database Systems to Maintain the Data Consistency
[研 究 者 ]
黃士銘; 羅偉丞
Huang, Shi-Ming; Lour, Weei-Cherng
[機構名稱  ]大同工學院資訊工程系(TTUINS)
[經費來源  ]行政院國家科學委員會(NSC)
[補助金額  ]204400
[補助編號  ]NSC87-2213-E036-003
[研究開始日期]1997/08
[研究結束日期]1998/07
[頁 冊 數 ]4頁
[語    文]英文
[建檔日期  ]1999/09
[出版情形  ]完成日期:民國87年9月
[原件提供  ]行政院國家科學委員會科學技術資料中心
NSC87-2213-E036-003
[科資分類號 ]IA0304
[關 鍵 詞 ]
異質性資料庫; 資料一致性; 交易管理; 資料庫管理系統; 動態資料庫; 動態
法則; 全域交易

Heterogeneous database; Data consistency; Transaction management;
Database management system (DBMS); Active database; Active rule;
Global transaction

[摘    要]
這篇論文提出了一個在動態異質性資料庫系統中的交易管理模式。為了完成交易的正確性與高效能,這個模式在全域階段提供一個以語意為基礎的平行控制機制,它擴展了依序式的執行模式,使得每個位置的資料庫能依不同順序來執行以提升效能。全域交易包含有資料交易和規則交易資料交易的處理機制是一般資料庫的交易管理,而規則交易的處理機制是在資料交易時,藉由動態規則的推論機來驅動。在全域交易處理期間,資料和規則交易藉由不同控制模組來管理。資料交易將會被分割成幾個子交易,全域交易模組將對各全域子交易產生一個正確平行排程而分送到各區段資料庫去執行。規則交易在規則推論期間將會產生新的全域交易。在全域階段提供語意基礎交易的好處是增強全域交易的效能。

This paper presents a framework to process the transactions
under the active heterogeneous database systems (AHDBS). In order
to perform a correct schedule of transactions with high performance,
the framework provides a semantic-based concurrency control on the
global level. It relaxes the correct criterion (global serializability) to allow the global subtransactions on each site to execute in different order. The global transaction includes data transactions and rule transactions. A data transaction is a general database transaction. A rule transaction is formed by active rules during the data transaction. During the global transaction process, both data and rule transactions are managed by different control
modules. The data transaction will be decomposed into several global sub-transactions. The global semantic based transaction module will produce a correct parallel schedule for the global sub-transactions. The rule transaction may generate new global transactions during the rule inference. The advantage of providing semantic-based transaction on global level will reduce the abort producing by the wrong global order of local database schedule and improve the performance of the global transaction.

[科資中心編號]RB8707-0269
[題    名]
超文件資料庫管理系統之研製
Design and Implementation of a Hypertext DBMS
[研 究 者 ]
劉寶鈞
Liu, Baw-Jhiune
[機構名稱  ]中央大學資訊工程系(NCUECSIE)
[經費來源  ]行政院國家科學委員會(NSC)
[補助金額  ]374800
[補助編號  ]NSC85-2213-E008-021
[研究開始日期]1995/08
[研究結束日期]1996/07
[頁 冊 數 ]68頁
[語    文]英文
[建檔日期  ]1999/10
[出版情形  ]完成日期:民國85年8月
[原件提供  ]行政院國家科學委員會科學技術資料中心
NSC85-2213-E008-021
[科資分類號 ]IA0304
[關 鍵 詞 ]
資料庫管理系統;超文件;物件導向;結構性查詢

Database management system (DBMS);Hypertext;Object orientation;
Structural query

[摘    要]
超文件系統(Hypertext system)的內部結構是由各種不同的資料節點
(Typed data nodes)和鏈結(Typed links)所組成的網路(Network),且節點與節點間由鏈結所連接。在瀏覽超文件的網路之前,可以先透過查詢節點的屬性以找到瀏覽的相關起始節點。結構性查詢(Structural query)則是透過結構的描述, 從一個超文件網路中取出包含所需的資訊的一組子網路。在論文中我們定義一個以圖樣(Pattern)為基礎之結構性查詢語言,並且提出查詢處理的方法。我們定義一組圖樣,且將查詢用圖樣來表示。在處理查詢上,我們將超文件系統中的節點(Nodes)與鏈結(Links)轉成物件導向資料庫(Object-oriented database)之物件(Objects)與聯結(Association)。如此以圖樣為基礎的查詢處理便可以轉成類似關連式資料庫所使用的關連代數(Relational algebra)的一組聯結代數
(Association algebra)。

The internal structure of hypertext systems is networks of
typed data nodes connected by typed links. Navigation in the network
of hypertexts could be done by first querying each local node and
then linking to relevant nodes according to node attributes.
Structural queries are queries that retrieve collections of
information from a network based on a specification of structures. In
this report, we define a pattern-based structural query language and
present query processing techniques. We define a set of patterns and
queries are expressed by patterns. In query processing, we map nodes
and links of hypertext applications into objects of an object-
oriented database. Objects and their associations in the object-
oriented database are uniformly represented by association patterns.
The pattern-based queries are then translated into a set of
association algebra which is analogous to the relational algebra used
for processing relational databases.


[科資中心編號]RB8705-0091
[題    名]
物件導向資料庫上的時程行為模型及查詢語言
A Behavioral-Temporal Object Oriented Database System---Object
Behavior Evolution Mechanism
[研 究 者 ]
陳國棟
Chen, Kuo-Tung
[機構名稱  ]中央大學資訊工程系(NCUECSIE)
[經費來源  ]行政院國家科學委員會(NSC)
[補助金額  ]369600
[補助編號  ]NSC85-2213-E008-022
[研究開始日期]1995/08
[研究結束日期]1996/07
[頁 冊 數 ]62頁
[語    文]英文
[出版情形  ]完成日期:民國85年8月
[原件提供  ]行政院國家科學委員會科學技術資料中心
NSC85-2213-E008-022
[科資分類號 ]IA0304
[關 鍵 詞 ]
物件導向資料庫;全時域資料庫;資料庫管理系統;查詢語言;演進歷史;事件軌跡

Object oriented database;Temporal database;Database management system
(DBMS);Query language;Evolution history;Event trace

[摘    要]
傳統資料庫著重於現時的資料,當新資料值產生後,舊資料值即被取代並丟棄,然而人們所關心的資料並不限於現時的資料,資料的歷史應該保存下來,許多研究已經著手於全時域資料庫系統(Temporal database),並有顯著的成果。行為支配著資料的變動,即有行為始有資料的變動,所以一個完整的全時域資料庫應該包含完整的資料演進歷史(Evolution histories)及事件軌跡(Event traces),資料演進歷史描述整個資料因事件發生而變化的歷史,而事件軌跡刻畫出整個事件的發生的來由。然而,現存的全時域資料庫系統對行為方面的討論付之闕如,以致無法從資料庫中獲得完整的資料演進歷史及事件軌跡,我們的目標即是在建立一個考慮行為的全時域資料庫的資料模式與查詢語言,由此資料模型,資料演進歷史及事件軌跡的結構能完整的描述出來並可由查詢語言讀取。本文提出一個物件導向資料模式,其利用有限狀態機用以描述資料演進歷史的結構,並藉由定義出物件程序間的呼叫關係,描繪出事件軌跡的樣式。文中並提出一個型樣式查詢語言,此語言可取出完整資料演進歷史及事件軌跡,並可容許不確定尋遊路徑(Uncertain navigation path)的情況下獲取資料。

Conventional databases concern itself with the most recent data.
As new data values become available through updates, the existing
values are replaced by the new data. However, the data which human
concern are not limited to recent data values. To meet the
requirement of temporal data including both past and current data,
traditional databases become insufficient. Many researchers have made
efforts toward achieving temporal databases that handle temporal
data.
Behavior affects data value, that is, behavior dominates the
change of data value. Therefore, a temporal databases should include
data evolution histories and event traces. Data evolution histories
describe how data evolve with events. Event traces depict the reason
why events take place. Nevertheless, existing temporal databases
concern nothing about behavioral aspect. They contribute to keeping
and retrieving different versions of data rather than complete data
evolution histories and event traces. This article aims at
constructing temporal data model and query language which incorporate
behavioral aspect. This data model depicts the structures of data
evolution histories and event traces. By using the query language
data evolution histories and event traces could be retrieved.
In this article, we construct a temporal data model, TORI model,
in which finite state machines are used to depict the structures of
data evolution histories. The structures of event traces are
described by defining the invocation relationships between methods of
objects. Besides, a pattern-based query language, BQL, is devised to
facilitate users to retrieve complete data evolution histories and
event traces from databases. This query language also avails to
handle uncertain navigation path.

[科資中心編號]RB8705-0076
[題    名]
多媒體資料庫系統之研究建立(二)
A Study on Multimedia Database System
[研 究 者 ]
張玉盈
Chang, Ye-In
[機構名稱  ]中山大學應用數學系(NCSCAPM)
[經費來源  ]行政院國家科學委員會(NSC)
[補助金額  ]527100
[補助編號  ]NSC85-2213-E110-034
[研究開始日期]1995/08
[研究結束日期]1996/07
[頁 冊 數 ]64頁
[語    文]英文
[出版情形  ]完成日期:民國85年8月
[原件提供  ]行政院國家科學委員會科學技術資料中心
NSC85-2213-E110-034
[科資分類號 ]IA0304
[關 鍵 詞 ]
同步控制;資料庫管理系統;循序性;利他鎖定法

Concurrency control;Database management system (DBMS);Serializability;
Altruistic locking method

[摘    要]
在資料庫系統中,要維持共享資料的一致性,就需要一個同步控制演算法(Concurrency control algorithm)去控制這些資料的存取動作。在先進式資料庫軟體中,比如說多媒體資料庫系統,傳統的循序性理論已經不足以滿足其需求。在另一方面,多媒體資料庫系統也需要擁有比關聯式系統更強大的模型技術。因此在這個計畫中,為了能在多媒體資料庫這種先進式資料庫軟體內增加同步控制的程度及改進系統效能,我們提出了一個以捐贈為基礎的物件導向資料庫同步控制協定。基本上我們所提出的協定是以ORION物件導向資料庫系統的鎖定方法及利他鎖定方法(Altruistic locking)為基礎而成。

In the database systems, maintaining the consistency of the
shared dates needs the concurrency control algorithms to controlling
accesses to these dates. In the advanced database applications, such
as multimedia database systems, traditional serializability theory is
not enough to meet the need of the advanced database applications. On
the other hand, a multimedia database system requires modeling
techniques more powerful than the ones offered by relational systems.
In this project, to increase the degree of concurrency control and
improve the performance of multimedia database systems, we propose a
donation-based concurrency control protocol for object-oriented
database systems. Basically, our proposed protocol is based on the
locking method in ORION object-oriented database system and
altruistic locking.

[科資中心編號]RA8604-2032
[題    名]
設計與實現一個在關連式資料庫管理系統上的物件導向包裝層
Design and Development of an Object-Oriented Shell for RDBMSs.
[研 究 者 ]
李強
LEE,CHIANG
[機構名稱  ]成功大學資訊工程研究所(NCKUIFEG)
[經費來源  ]行政院國家科學委員會(NSC)
[補助金額  ]190000
[補助編號  ]NSC84-2213-E006-033
[研究開始日期]1995/02
[研究結束日期]1995/07
[頁 冊 數 ]65頁
[語    文]中文
[建檔日期  ]1997/11
[出版情形  ]完成日期:民國84年10月
[原件提供  ]行政院國家科學委員會科學技術資料中心
NSC84-2213-E006-033
[科資分類號 ]IA0304
[關 鍵 詞 ]
關聯式資料庫;物件導向資料庫;資料模型;資料庫管理系統

Relational database;Object-oriented database;Data model;Database
management system (DBMS)

[摘    要]
物件導向資料庫管理系統(Object-oriented database management
systems)已經被證明在某些領域的應用有較佳的評價,例如:CAD(Computer aided design)、GIS(Geography information systems)、OIS (Office information systems)、MULTIMEDIA等,但是,在過去的若干年裡,由於關連式資料庫管理系統(Relational database)的盛行,大量的資料已經以關連式資料模式建立於關連式資料庫管理系統。而今,這些資料成了欲使用OODB的障礙了。因此,本計畫之宗旨,是要在一個關連式資料庫系統之上加上一個Object-oriented shell,使這個關連式資料庫系統(RDBMS)兼具有關連式資料模式及物件導向資料模式的特性。以使物件導向資料庫的使用者,可以透過OOSR的轉換,而使用建立於關連式資料庫系統的資料。我們所實現的OOSR系統,包括五個組模---Query Trans、Schema Trans、Task Handler、Schema Info以及OOSR Engine。其中,Query Trans將接收所有由使用者所給予的Query,根據Schema Info中所提供有關Schema的資料,將使用者依物件導向架構所給予的Query,轉換為可在關連式資料庫執行的Query;SchemaTrans是一個提供給資料庫管理者(DBA,Database administrator)使用的工具,透過這個工具,DBA可將已建立的關連式資料架構轉換為物件導向資料架構;Task Handler將依Attributes subquery及Method subquery產生執行排程(Scheduling);Schema Info 是一個使用者不可視的一個模組,負責記錄OOSR中,使用者所定義的物件導向架構的所有資訊;OOSR Engine模組是負責處理方法子查詢,並且將子查詢結果送回Task Handler中。

It has been widely recognized that object-oriented database
management systems are suitable for applications of complex data
types and relationships. The applications include CAD/CAM, GIS,
multimedia, OIS, etc. On the other hand, due to the sound
mathematical support and the internationally standardized query
language, relational database management systems have attracted great
attention from various applications. Moreover, many real databases
have been build into relational databases. These data, however,
become an obstacle for being further applied to OO world, resulting
in a limited use of these existing relational data for resolving
problems. The main focus of this project is to design and develop an
Object-Oriented Shell for Relational Databases (OOSR) such that OO
users (application programs) can also access those relational data.
This topic involves advanced research and development and is a
frontier work at this moment.
Four major tasks have been accomplished in this project. (1)
Query Trans: This is a module in OOSR which is responsible for
translating user query (in OO form) into a relational query. (2)
Schema Trans: This module is designed to translate the user defined
OO Schema of data into an equivalent relational schema. Also, it is
used for translating the resultant relational data back to OO data.
(3) Schema Info: This module provides the schema information of the
underlying relational data and the conversion information of OO
schema to relational schema. It is used for both query and schema
translations. (4) Task Handler: This module coordinates, schedules,
and controls the work of all other modules. (5) OOSR Engine: This
module is responsible for processing the method sub-queries and
sending back the query result to the Task Handler.

[科資中心編號]RB8707-0172
[題    名]
多媒體資料庫的查詢排程最佳化
Query Scheduling Optimization for Multimedia DBMS
[研 究 者 ]
王熹偉
Wang, Shi-Wei
[機構名稱  ]交通大學資訊工程系(NCTECSIE)
[經費來源  ]行政院國家科學委員會(NSC)
[補助金額  ]266500
[補助編號  ]NSC84-2213-E009-044
[研究開始日期]1994/08
[研究結束日期]1995/07
[頁 冊 數 ]31頁
[語    文]英文
[建檔日期  ]1999/10
[出版情形  ]完成日期:民國86年3月
[原件提供  ]行政院國家科學委員會科學技術資料中心
NSC84-2213-E009-044
[科資分類號 ]IA0304
[關 鍵 詞 ]
多媒體;資料庫管理系統;查詢排程最佳化;即時存取;同步化

Multimedia;Database management system (DBMS);Query scheduling
optimization;Realtime access;Synchronization

[摘    要]
由於多媒體系統整合多種不同的媒體例如文字、聲音、影像等,所以能夠讓使用者擁有良好的人機介面,但其最後成功關鍵仍取決於各式各樣的支援技術,其中之一是即時性的多媒體資料庫管理系統,它的主要功能在於能夠管理不同種類的資料,藉由即時性多媒體資料庫管理系統,使用者可以在有效時間內存取想要的資料並播放如聲音、影像等連續性的資料。

Because of the integration of various representing media such
as text, audio, video, image, etc., multimedia systems can provide
the most natural way for people to communicate with computers.
However, the real success of multimedia systems relies on the
maturity of various supporting technologies. One of the most
important technologies is real-time multimedia database management
systems (DBMS). The main function of real-time multimedia DBMSs is to
manage various kinds of data. Via the management of the multimedia
DBMSs, users can play back continuous data such as audio and video
and access the desired information from databases within reasonable
response time.

[科資中心編號]RA8604-2112
[題    名]
使用者階層上關係值的計算之研究
The Study of Computing the Relationship between Two Users in a
Hierarchy.
[研 究 者 ]
詹進科
JAN,JINN-KE
[機構名稱  ]中興大學應用數學系(NCHSAPM)
[經費來源  ]行政院國家科學委員會(NSC)
[補助金額  ]204000
[補助編號  ]NSC84-2213-E005-001
[研究開始日期]1994/08
[研究結束日期]1995/07
[頁 冊 數 ]33頁
[語    文]英文
[建檔日期  ]1997/11
[出版情形  ]完成日期:民國84年7月
[原件提供  ]行政院國家科學委員會科學技術資料中心
NSC84-2213-E005-001
[科資分類號 ]IB0002
[關 鍵 詞 ]
資料庫管理系統;使用者階層;資料庫管理師;資料庫建構者;一般使用者

Database management system (DBMS);User hierarchy;Database
administrator;Database creator;Casual users

[摘    要]
在資訊安全系統中,使用者之間可能因為各自代表不同的職務而在計算機內部有某一特定關係值,用來定義彼此之間的階層關係。另一方面,藉著這個關係值,系統可以決定使用者之間檔案等資源的存取權。用一般的資料庫管理系統(DBMS)來說,資料庫之使用者可分成三個階層:資料庫管理師(DBA)、資料庫建造者(DBC)及一般使用者(CU)。將DBA定義成使用者階層樹狀結構圖(User hierarchy tree)上之最上層樹根節點,用於表示其最高權力之主體。其下再配置DBC,最後在DBC之下配置所有使用者以完成整體架構。使用者階層關係圖可以用很多種資料結構來表示。然而何者為佳則取決於:(1)使用者間之關係值可以經過簡易計算而得,(2)階層圖在計算機內部之儲存空間相對地經濟,以及(3)使用者間之關係值不易被侵入者偷窺或竄改。在本計畫之成果報告中,我們提出(1)一種轉換方法以改善階層圖之儲存效率,(2)一種演算法以最適化使用者之鍵值,(3)一個機制以計算出使用者間之關係值。總結而言,我們的研究結果達成了以下五項優點:(1)藉著轉換方式以增加資訊安全度;(2)系統僅存放使用者鍵值以防儲存空間之浪費;(3)任意使用者間之關係值可以經簡易計算而得;(4)最大鍵值可以達到最適化目標;(5)我們所提的方法能夠應用於所有機器上。

In an information protection system, the relationships of users may be represented by some internal forms which are devised based on the hierarchy of users. On the other hand, the O.S. may determine the
access right of one user to another based upon the corresponding
relationship. In a general Data Base Management System (DBMS), three
levels of users are involved; the first is the Data Base Administrator (DBA), the second is the Data Base Creators (DBC) and the third is the Casual Users (CU). All of these users are placed on the user hierarchy tree accordingly in which the DBA on the root and the DBC are placed as DBA's children and the CU are as their children recursively.
Quite a few of data structures may be used for representing the user hierarchy. The criteria of good scheme include (1) the relationship of user hierarchy may be computed effectively and easily, (2) the storage space for restore the said user hierarchy tree is rather small or economic and (3) the relationship between any two users is not possibly wiretapped or modified by any intruder. Our proposal in this study includes (1) designing a transformation scheme to improve the efficiency for storage space, (2) designing an algorithm to optimize the key values and (3) given a mechanism to reveal the relationship between any two users. In summary, our proposal achieves (1) the security is enhanced with the reordering (transforming) process; (2) the system stores only some key values to prevent wasting space; (3) the relationship between any two users can be easily revealed through simple arithmetic operations; (4) the maximal value of all key values is
reduced to an optimal level; (5) the proposed scheme can be
implemented on any machine.

[科資中心編號]RA8504-2055
[題    名]
超文件資料庫管理系統之研製
Design and Implementation of a Hyper-Text DBMS.
[研 究 者 ]
劉寶鈞
LIU,BAW-JHIUNE
[機構名稱  ]中央大學資訊及電子工程研究所(NCUGCSEG)
[經費來源  ]行政院國家科學委員會(NSC)
[補助金額  ]284000
[補助編號  ]NSC83-0408-E008-017
[研究開始日期]1993/08
[研究結束日期]1994/07
[出版日期  ]1994/08
[頁 冊 數 ]75頁
[語    文]英文
[建檔日期  ]1997/01
[出版情形  ]研究報告,75頁,民國83年8月
Report, 75p., Aug. 1994
[原件提供  ]行政院國家科學委員會科學技術資料中心
NSC83-0408-E008-017
[科資分類號 ]IA0304
[關 鍵 詞 ]
物件導向資料庫;超文件;Smalltalk系統;資料庫管
理系統

Object oriented database (OODB);Hyper text;Smalltalk system;Database
management system (DBMS)

[摘    要]
超文件系統的核心為資料庫管理系統,但
超文件處理的是大物件,不同於目前一般資料
庫系統處理的文數字資料,這些文數字資料稱
為小物件,通常均為結構化的資料,大物件則屬
非結構化的資料,如文件、影像、語音訊號等
。大家均公認物件導向資料庫系統最具有處理
多媒體資訊(包括大物件極小物件)的能力,但目
前發展的物件導向資料庫系統,尚未具備這種
能力,因此本研究乃從超文件資料庫處理特性
著手,先界定所需的基本運算及存取大物件的
方法,進而採用直接修改及擴充SMALLTALK系統,建
立一個適合超文件系統使用的物件導向資料庫
管理系統。

DBMS is the kernel of a HyperText system. This DBMS differs
from a conventional DBMS which manipulates and manages
alphanumerical data (called small object), it deals with large
objects. Usually, small objects are structured data while large
objects are unstructured data. For instance, text, image, and voice
signal are large objects. We are convinced that an object-oriented
database system is a potential system for multimedium data (i.e. a
mixed data with large objects and small objects), yet there is still
no such system which can fully support this solution. In this
research, we start from investigating the characteristics of the
manipulation in HyperText system to identify the primitive
operations. Then we also study the access methods for large objects.
After that, we design and implement an OODBMS by modifying and
extending the Smalltalk virtual machine.

[科資中心編號]RA8404-2065
[題    名]
交談式多媒體資料處理系統之設計與建立
The Design and Implementation of an Interactive Multimedia Information
Handling.
[研 究 者 ]
劉德明
LIU,DER-MING
[機構名稱  ]逢甲大學資訊工程系(FCUEIFE)
[經費來源  ]行政院國家科學委員會(NSC)
[補助金額  ]273000
[補助編號  ]NSC83-0408-E035-002
[研究開始日期]1993/08
[研究結束日期]1994/07
[出版日期  ]1994/09
[頁 冊 數 ]86
[語    文]英文
[建檔日期  ]1996/01
[科資分類號 ]IA;IA0304
[關 鍵 詞 ]
多媒體資訊系統;資料庫管理系統;伺服器;委託 者

Multimedia information system;Database management system;Server; Client

[摘    要]
在現今的社會中,多媒體的應用是越來越 受歡迎了,多媒
體使得人們更容易接受電腦並 且將電腦使用的更好。一個
交談式的多媒體系 統包括應用程式以及資料庫。使用資料
庫不僅 可以增加程式設計師的產量並且使系統具有一 致性
也更容易維護。我們設計並實作一個交談 式多媒體資訊系
統,其中包含兩個子系統:物件 導向的多媒體資料庫系統和多
媒體展示系統, 皆實作在主從架構的環境中。
物件導向的多媒體資料庫系統負責管理多 媒體物件,傳
送多媒體物件到多媒體展示系統 以供展示,並且提供時間軸
編輯器以方便在時 間軸上編輯多媒體物件。在真實世界中
每個個 體皆可被視為物件,我們也以這種概念將交談 式多媒
體資訊系統模型化。每個媒體像是視訊 、聲音、影像以及
文字等,在我們系統中皆被 視為物件。我們也為系統中的各
種多媒體物件 它們的模型,同時設計群組物件模型以管理並
控制多媒體物件的同步關係。
多媒體展示系統是用來展示多媒體物件, 它是運用OCPN同
步模型來維繫各個多媒體物件 之間的同步關係。為了容易
編輯和得到較佳的 同步效果,我們也提供了同步模型和時間
軸之 間的轉換器以轉換成彼此間的結構。另外,在 第一階段
的計畫中,我們對於視訊和聲音間的 精密同步提出解決方案,
在本期計畫中,我們將 此方法實際的運用在本系統中,有不錯
的效果, 我們更加以修正,以期達到更好的境界。在主 從架
構的環境中使用本系統,可以讓使用者輕 易的發展交談式多
媒體資訊處理的應用程式。

The application of multimedia is becoming more popular nowadays.
Multimedia can make people more receptive to computers and make a better use of them. An interactive multimedia system contains application programs and databases. Using databases will not only increase the productivity of application programmers, but also produce a more consistent system which can be easily maintained and may evolve with much less programming effort. We have designed and implemented an interactive multimedia information handling system which consists two subsystems: the object-oriented multimedia database system and the multimedia presentation system. These two subsystems are implemented in a client-server environment via network connection.
The multimedia database system is capable of managing multimedia
objects, delivering multimedia objects to the multimedia presentation system for presentation, and providing a timeline editor to edit multimedia objects in timeline. Since everything can be treated as an object in the real world, we use the object concepts to model our system. Every media, such as video, audio, image, and text, is an object in our system. We design a multimedia object model for each one of them, and a group object model to handle the synchronization. The multimedia presentation system is used to present multimedia objects with the OCPN synchronization model. For easy editing and good synchronization performance, we also provide a timeline-OCPN converter to convert the timeline structure to and from the OCPN structure. In addition, a fine-grain synchronization mechanism is also provided. Using this infrastructure in a client- server environment, users can easily develop their distributed multimedia information applications.

[科資中心編號]RA8304-2042
[題    名]
多媒體資訊網路系統之研製
Developing a Multi-media Information System on a Local Area Network.
[研 究 者 ]
黃國禎
HUANG,KUO-CHENG
[機構名稱  ]交通大學計算機中心(NCTUCC)
[經費來源  ]行政院國家科學委員會(NSC)
[補助金額  ]351000
[補助編號  ]NSC82-0408-E009-049
[研究開始日期]1992/08
[研究結束日期]1993/07
[出版日期  ]1993/07
[頁 冊 數 ]61
[語    文]中文
[建檔日期  ]1994/12
[科資分類號 ]IA;IA0304;IA0202
[關 鍵 詞 ]
多媒體系統;資訊系統;資料庫管理系統; 計算機網路;區域網


Multi-media system;Information system; Database management system
(DBMS);Computer network;LAN

[摘    要]
本計畫主要的目的在於研究一套多媒體資 訊網路系統
。目前該資訊系統中包括了四部終 端設備及一個管理系統
。每部終端設備都提供 了一個多媒體的使用者介面。管理
系統中使用 TCP/IP協定的區域網路負責系統間的的溝通,主 控
系統負責監控終端機的所有行為,一個物件 導向的資料庫管
理系統用來維護所有的多媒體 檔案,劇本檔產生系統產生控
制終端機行為的 劇本檔。除此之外,尚有一個預先處理系統
做 影像檔或動畫檔與中文字型的合成,目前這套 資訊管理系
統已被國立交通大學採用。系統提 供有關校園生活的各種
資訊,內容包括學校組 織、學校中提供的各種軟硬體設備、
如何使用 這此資源、如何獲得各種服務的指式,交通的 資訊
,對各系所的介紹,目前進行中計畫的簡介, 校園活動、電影
及各種活動的宣佈等。藉著網 路的通訊設備,有效率的管理
這些多媒體資訊 必須引用一些新的技術在管理系統的設計
上。 經過562份對使用者的有效問卷的統計中可發現 資訊網
路系統的實施是成功的。未來在實際應 用上,我們更將擴展
至二十部以上的終端設備 進行連線操作。

In this project, we develop a campus information network system, which
consists of several terminal devices and a management system. For each
terminal device, a multi-media user interface is provided, and a local area
network TCP/IP protocol is employed to enable the communications from the management system. The management system consists of main control system for monitoring the behaviors of the terminal devices, a object-oriented database management system to maintain media-media data files, a script generating system to produce control scripts for terminal devices, and a preprocessing system to compose images or animations with Chinese fonts. The information system has been installed in the National Chiao Tung University of R.O.C. to provide all kinds of information concerning campus life, including the organization of the school, available software and hardware resources, instructions of how to apply the usage of a resource, instructions about all kinds of available services, traffic information, introductions for each institute and department, introductions for on-going projects and researches, campus activities, campus movies, all kinds of announcements, etc. To efficiently manage the multi-media data through network communications, some new techniques are applied to the design of the management system. From the statistic results by querying 562 users, we conclude that the whole project is successful.

[科資中心編號]RA8410-2023
[題    名]
以知識庫為基底之語意查詢最佳器之研究
Knowledge-Based Approach for Semante Quary Optimization.
[研 究 者 ]
金陽和
CHIN,YEH-HAO
[機構名稱  ]清華大學資訊科學研究所(NTHUCOS)
[經費來源  ]行政院國家科學委員會(NSC)
[補助金額  ]259000
[補助編號  ]NSC81-0408-E007-524
[研究開始日期]1992/02
[研究結束日期]1993/01
[出版日期  ]1993/04
[頁 冊 數 ]62頁
[語    文]中文
[建檔日期  ]1996/06
[出版情形  ]研究報告,62頁,民國82年4月
Report, 62p., Apr. 1993
[科資分類號 ]IA0304;IA0407
[關 鍵 詞 ]
知識庫系統;語意查詢;資料庫管理系統

Knowledge-based system;Semantic quary;Database management system

[摘    要]
語意查詢最佳化是使用資料的知識,例如
整體限制,將原始的查詢轉換成一個新的查詢,
這個新的查詢是與原始查詢語意等價的。而且
被執行時將會比原始查詢有更好的效率。在本
論文中,我們提出一個系統叫KBSQP(以知識庫為
基底之語意查詢最佳器),它是一個以知識庫為
基底的方法來處理語意查詢最佳化, KBSQP使用
整體限制庫來儲存整體限制和最佳化技術知識
庫作查詢之最佳化。我們提出了一個有效率的
機構來選擇與合併相關的整體限制到查詢處理
中來產生語意等價的查詢處理。在KBSQP中的最
佳化技巧是用規則的形式來表現,而且是量化
的。我們的目的是提出一個具有資料庫獨立性
之語意查詢最佳器,而不是僅僅一群高階層的
指引去作語意查詢之處理。

Semantic query optimization uses knowledge about data (e.g.,
integrity constraints) to transform a query into a form that is
semantically equivalent to original query and being likely to
answered in more efficiently than the original one. In this thesis
we propose a system called KBSQP (Knowledge-Based Semantic Query
Processor) which is knowledge-based approach to semantic query
optimization. KBSQP uses an integrity constraint base to store
integrity constraints and an optimization technique knowledge base
to optimize query. An efficient mechanism is developed to select and
merge the relevant integrity constraints with a given query for
generating the semantically equivalent queries. An optimization
technique of KBSQP is represented as a rule which is quantified, and
presents a DBMS independently semantic query processor, rather than
a set of high-level guide-lines for semantic query processing.

[科資中心編號]RA8304-2033
[題    名]
資料庫管理系統與專家系統的整合
An Integration of Database and Expert Systems.
[研 究 者 ]
陳秋華
CHEN,CHYOU-HWA
[機構名稱  ]台灣工業技術學院電子工程技術系 (NTITENT)
[經費來源  ]行政院國家科學委員會(NSC)
[補助金額  ]347000
[補助編號  ]NSC81-0408-E011-510
[研究開始日期]1992/02
[研究結束日期]1993/01
[出版日期  ]1993/10
[頁 冊 數 ]128
[語    文]中文
[建檔日期  ]1994/12
[科資分類號 ]IA;IA0304;IA0407
[關 鍵 詞 ]
資料庫管理系統;專家系統;整合

Database management system (DBMS);Expert system;Integration

[摘    要]
為了使專家系統在其知識庫極為龐大時仍 然能有效運
作,在另一方面也可使資料庫系統 具有推理的能力,所以整合
專家系統和資料庫 系統有其實際的需要。建立此新的專家
資料庫 系統(Expert database system),我們將專家系統中的 事實
(Facts)改以資料庫的方式儲存,以解決資料 量大時記憶體空間
不足的狀況。在這種環境 下,我們選擇TREAT演算法作為Rule的
比對的方法, 因為TREAT比對演算法的效率優於其他的比對演
算法,而且TREAT比對演算法非常適合用DBMS的方 式來實行。為
因應DBMS的處理需要,我們對TREAT 比對演算法做適當的修改,並
且提出一個新的 比對架構。在比對過程中的所有部分比對
(Partial match)的結果,以及比對完成後所得的結 果,我們亦以資
料庫的方式儲存,以達成專家系 統和資料庫系統全面性整合
的目的。但由於傳 統Sequential rule firing之執行方式對磁碟做
I/O的 次數頻繁,所以系統的效率並不好。為了減少 I/O的次數
,我們在此應用了Parallel rule firing理論 的分析結果使得此新
系統的效率顯著提昇。經 由數個實驗數據的結果證明,應用
Parallel rule firing理論後,在平行性愈高的應用程式,且資料 量
愈大時,此系統的執行效率會愈高。

Intergratrion of expert systems and database systems promises to make
expert system work efficiently when it needs a very large knowledge base, while endowing database systems with inferencing capability. We designed and implemented a system, called TIGER, that is a total integration of expert systems and database systems. In our system, all the facts of in traditional expert systems are stored in the relational database component of TIGER. All partial matchs, intermediate facts, and rule instantiations are also stored in the temporary database. To execute the rules of expert systems, we chose the TREAT match algorithm, appropriately modified, as a basis for rule matching because of its superior efficiency in comparison with other match algorithms and its suitability for integration with a DBMS. However, traditional rule execution strategy is sequential, resulting in extremely high frequency of disk access, and low overall system performance. We propose a novel and vital idea in our system to significantly reduce the number of joins by applying parallel rule firing theory to rule execution in expert database systems. The intuition behind the idea is that when multiple rules are discovered to be firable in parallel, their joins can be performed at the same time, instead of being done separately, as traditional sequential execution of rules demands. Our experimental results confirms our intuition and demonstrates that as more parallel rules are present in the application, and as the data set gets larger, the system achieves very impressive performance.

[科資中心編號]RA79-2116
[題    名]
法則式專家系統與關聯式資料庫管理系統耦合之研究
A Study of Coupling the Rule-based Expert System and the Relational Database
Management System.
[研 究 者 ]
曾怜玉
TSENG,LIN-YU
[機構名稱  ]中興大學應用數學研究所(NCHYAPMG)
[經費來源  ] 行政院國家科學委員會(NSC)
[補助金額  ]229000
[補助編號  ]NSC79-0408-E005-01
[研究開始日期]1989/08
[研究結束日期]1990/07
[出版日期  ]1990/07
[頁 冊 數 ]30
[語    文]英文
[建檔日期  ]1998/07
[出版情形  ]IASTEDInternationalConferenceon
Artificial Intelligence Applicati!!ons and Neural Networks,p.25-27,1990.
[原件提供  ]行政院國家科學委員會(NSC)
微縮片號: NSC79-0408-E005-001 1片 39幅
[科資分類號 ]IA;IA0407
[關 鍵 詞 ]
專家系統;資料庫管理系統;耦合

Expert system;Database management system;Coupled system

[摘    要]
本計畫預期要把一個專家系統與一個資料庫管理系統耦合
在一起,在本報告中,我們提出一種鬆耦合的方法,可以成功地
把一法則式專家系統與一關聯式資料庫系統耦合起來,使得
其應用的範圍與能力大增.我們的方法主要是利用專家系統
內的推論機來幫我們把對資料庫諮詢之程式組合好,並幫我
們把此程式傳給資料庫管理系統,由於利用了推論機,所以耦
合的介面非常容易建立,並不需寫很多程式,此介面包括了三
部份:ぇMeta Datatase(MDB),用來描述資料庫.えMeta Knowledge Base(MKB),包含一些法則,用來組成對資料庫諮詢之程式.ぉExternal
functions,用來組合queries,把程式傳給資料庫管理系統.我們也
做了實驗,證明這方法是相當容易且有效的.

In this report, we propose a simple scheme to couple an expert system and a database management system on personal computers. There are many database management system and expert system building tools available on personal computers. It is quite useful if we have a simple way to couple these systems to fulfill the requirements of applications. Since there is an inference engine in the expert system, the main idea of our approach is to make use of this inference engine to help us composing the queries and then passing the queries to the database management system. In building the interface between the expert system and the database management system, we first build the meta database (MDB), which is a description of the database in the coupled system. In many cases, MDB may be as simple as a table, so it is easy to construct. Secondly, we build the meta knowledge base (MKB), which consists of the knowledge needed to help us composing the queries to DBMS. Finally, we need several external functions, which do the job of passing queries from ES to DBMS and passing answers from DBMS to ES. The inference engine together with MDB, MKB and external functions constitute the interface of ES and DBMS. Some experimental results are given. This experiment shows that our scheme is very simple and effective.


[科資中心編號]RA78-2109
[題    名]
物體導向多媒體資料庫系統之研究
A Study on Object-Orinted Multi-media Database Systems.
[研 究 者 ]
楊維邦
YANG,WEI-PANG
[機構名稱  ]交通大學資訊科學系(NCTSIFS)
[經費來源  ]交通部電信總局電信研究所(TCLG)
[補助金額  ]629800
[補助編號  ]C77057
[研究開始日期]1988/04
[研究結束日期]1989/03
[出版日期  ]1989/03
[頁 冊 數 ]175
[語    文]中文
[建檔日期  ]1998/07
[科資分類號 ]IA;IA0304
[關 鍵 詞 ]
個體導向;多媒體資料庫系統;資料庫管理系統

Object-orinted;Multi-media database systems;Database management system

[摘    要]
傳統的資料庫系統的資料模型(data model)對於商業上的一般
文字資料可以處理的很好,但是由於資料庫系統的應用範圍
日益闊,它所管理的資料不僅限於文字而已,更包括了圖形,影
像及聲音等各種型態的資料.一個能統合管理這些不同型態
資料的多媒體資料庫系統便應運而生了,然而使用傳統的資
料模型來描述這些不同型態的資料並不適合.因此本計畫在
研究以個體導向(object-oriented)資料模型來發展多媒體資料庫
系統的方法,因為個體導向資料模型有較強的描述能力,可以
較簡單而直接的以一個統一的觀念,來描述各種不同型態的
資料,適合於發展多媒體資料庫系統.
本計畫期末報告將對這一年來所做的研究做一總結,首先
對於個體導向資料模型的觀念做一完整的描述.接著對個體
導向資料庫系統上的協同控制(Concurrency control)的問題做一深
入的探討,並提出我們研究的心得.最後在附錄中介紹我們以
Smalltalk,在PC上製作的"校園資訊系統"的Front-end,並附上我們所
蒐集有關個體導向資料庫系統的詳盡參考資料,提供做為更
進一步研究的基石.此外附上一篇曾發表於"電腦季刊"上的"
淺談多媒體資料庫系統",可做為初次接觸多媒體資料庫系統
的入門參考資料.

The data models of traditional database management systems (DBMS) are very suited for handling the word data in the commercial applications. But the application domains of DBMS are becoming more and more broad. The data types managed by DBMS include not only word data but also graphics, image and sound etc. A multimedia DBMS that can handle these different kinds of data types will come for this, but it is not suitable to use traditional data models to describe these different kinds of data types. In the research plan, we research the methods of using object-oriented data model to develop a multimedia DBMS. Because the object-oriented data model has good ability of description and can use a integral concept to model different kinds of data types, it is suited for developing a multimedia DBMS. In this final report, we make a conclusion of research work that we did in this year. At first, we describe the concepts of object-oriented data model completely. Then we will show some results on the problems of concurrency control in the object-oriented DBMS. At the appen dix, we introduce the front-end of the "Compus information system" that we implement on PC using Smalltalk-80, and show the complete reference papers about object-oriented DBMSs that we collected these years. They can be used as the foundation for deeper researches. At last, we show an article-"Introduction to object-oriented DBMS" that can be used as a very good reference for those who want to touch the field of object-oriented DBMSs.

[科資中心編號]R75-TK7885-040
[題    名]
關聯型資料庫管理系統之設計
Design and Implementation of a Relational Database Management System.
[研 究 者 ]
劉寶鈞; 陳國棟
LIU,BAW-JHIUNE ; CHEN,GWO-DONG
[機構名稱  ]中央大學(NCUG)
[經費來源  ]行政院國家科學委員會(NSC)
[補助金額  ]199000
[補助編號  ]NSC74-0408-E008-04
[研究開始日期]1984/12
[研究結束日期]1985/11
[出版日期  ]1985/12
[頁 冊 數 ]211
[語    文]中文
[建檔日期  ]1987/06
[原件提供  ]行政院國家科學委員會(NSC)
微縮片號: NSC74-0408-E008-004 3片223幅
[科資分類號 ]TK7885
[關 鍵 詞 ]
關聯型資料庫; 資料庫; 資料處理; 系統程式

Relational database ; Database ; Data processing ; System programming

[摘    要]
本文描述一個關聯型資料庫管理系統RDBMS的資料語言處理
系統的設計與製作.此系統提供兩種語言:關聯代數與SQL* c關
聯代數除提供傳統的詢訊功能外,本系統增加資料處理,資料
定義及控制等功能,使它成為獨立的資料語言.SQL*語言係改進
SEQUEL 2(SQL的前身)的缺點而設計的資料語言.本系統增加其
定義域的定義指令及完整性約束的外鍵宣告指令,使它提供
的關聯架構更加符合關聯資料庫系統的需求.本系統亦提供
鉗入式SQL*的使用,使用者可以利用它在C語言之程式中存取資
料庫的資料.

This report describes the design and implementation of the data sublanguage processing system in a Relational Data Base Management System RDBMS The system supports two kinds of languages: relational algebra and SQL*. In addition to the query function, the relational algebra is enhanced with the data manipulation, data definition and data control facilities, thus making it a complete data sublanguage for a relational data basesystem. SQL*, originated from SEQUEL 2, is extended with the domain definition and foreign key declaration statements to meet the requirements of relational data base systems. The system also supports embedded SQL* facilities in C program. One can access the database through his C program by means of this facilities.

[科資中心編號]R73-TK7885-054
[題    名]
一個分散式資料庫管理系統之模擬系統之設計與製作
The Development of a Simulation System for a Distributed Database
ManagementSystem.
[研 究 者 ]
羅德和
LO, DER-HER
[機構名稱  ]交通大學資訊科學系(NCTSIFS)
[經費來源  ]行政院國家科學委員會(NSC)
[補助金額  ]0
[補助編號  ]NSC72-0404-E009-R16
[研究開始日期]1983/12
[研究結束日期]1983/09
[出版日期  ]1983/10
[頁 冊 數 ]114
[建檔日期  ]1998/07
[出版情形  ]

[原件提供  ]行政院國家科學委員會(NSC)
微縮片號: NSC72-0404-E009-R16 2片 123幅
[科資分類號 ]IA
[關 鍵 詞 ]
查詢處理;電腦模擬;資料庫管理系統
Query processing;Computerized simulation;Database management system
[摘    要]
分散式資料庫系統之查詢處理問題,是探討分散式資料庫
系統領域內一個尚待深入研究之主題。本篇論文的目的即是
探討此此問題,以尋找一個實用,有效率的分散查詢處理策略
。!!在以處理查詢時所需之通訊成本為主要考慮因素下,本文
提出了四種新的查詢處理策略。利用數學模式與模擬方法,
以分析,比較各種處理策略之得失,進而推論各種策略的適用
環境。並將其中一個較適合者,製作在一個現存於VAX-11/780VMS
作業系統上之DRDBMS系統上。!!另外與查詢處理有關之問題,如
代數查詢運算結果記錄數的預估技巧,分散式資料庫管理系
統之結構,分散查詢處理之作業控制等,本文亦有詳盡之討論

This paper addresses query processing in a distributed relational
databasemanagement system. It intends to find an effective distributed query processingstrategy. Assume the communication cost is the only cost of query processing.This paper proposes four distributed query optimization algorithms. Mathematicalmodeling and computer simulation are employed to analyze, compare these fouralgorithms. The applicable environment for each algorithm is concluded.One of the algorithms is implemented on a distributed relational detabasemanagement system which can run on VAX/VMS operating system. Besides, theestimation technique for relational operations, the architecture of thedistributed relational database management system, and the scheduling controlof the distributed query processing are also discussed.



[科資中心編號]R73-TK7885-056
[題    名]
多模式資料庫管理系統之研製
Desing and Implementation of a Multi-Model Data Base Management System.
[研 究 者 ]
龐台銘;劉寶鈞
PARNG, TAI-MING;LIU, PAO-JIUN
[機構名稱  ]台灣大學電機工程系(NTUEELE)
[經費來源  ]行政院國家科學委員會(NSC)
[補助金額  ]151200
[補助編號  ]NSC72-0404-E002-04
[研究開始日期]1982/10
[研究結束日期]1984/03
[出版日期  ]1984/04
[頁 冊 數 ]125
[建檔日期  ]1998/07
[原件提供  ]行政院國家科學委員會(NSC)
微縮片號: NSC72-0404-E002-04 2片 132幅
[科資分類號 ]IA
[關 鍵 詞 ]
多模式資料庫系統;資料庫管理系統;交談方式;多模式資料庫
系統;多模式資料庫系統;多模式資料庫系統

Interactive;Database management system;Multi-model database system

[摘    要]
目前的資料庫系統主要是根據網狀式,階層式,及關係式三
種資料模式而建造。由於這些不同模式的資料庫系統並存,
造成不同系統的使用者難以溝通,使用者訓練和教育困難,系
統之間程式和資料無法有效轉移,以及系統不易更換等問題
。所以有必要建立一個同時能使三種資料模式並存的資料庫
系統。!!關係式資料庫系統,由於理論基礎堅強,觀念簡單,使
用方便等優點,已成為目前建立資料庫系統的主要趨向。但
是,經由實際應用發現,關係式資料庫系統對於諸如電腦輔助
設計系統中的圖型影像,辦公室自動化系統中的表格與文書
資料等有類於階層式或網狀式的資料較不易描述,而亟需有
所改進。!!基於以上二方面的要,求我們設計一個以關係式資
料庫系統為基礎而對同一資料同時提供關係式,階層式,網狀
式三種觀點的多模式資料庫系統。在這個多模式資料庫系統
上,使用者介面是交談式的通用資料庫語言,以單一語言結構
即可對同一資料的各種模式觀點操作。!!在台灣大學電機系
系的VAX-117/780計算機系統上,已經建造完成一個關係式資料庫
系統。多模式資料庫系統的製作方法是就關係式資料系統加
以擴充改進,以VAX-11FORTRAN為撰寫語言,並儘量利用VAX/VMS作業系統和RMS檔錄管理系統等系統軟體設施。

On account of its promised feature, the relational data model is
oftenspeculated to be the best candidate for the basis of database systems.
Yetexperiences reveal that relation database systems are not always suited
forhandling some kinds of data such as images, text, messages, and so on.
Thesehetergeneous data are in essence conformable to data models other than therelational data model. Hence we contrive a multi-model database system whichsupports the three major data models, namely, the network, the hierarchical,and the relational data models. The design issue is that this multi-modeldatabase system such that, while preserving the virtue of the relational datamodel, it can simultaneously offer the user the other two
conceptual pointsof view upon the same piece of data. In addition, the sole user interfaceis achieved via a unified standalone data language
whichoperations across the three data models with the identicallanguage
contents.!!The proposedmulti-model database system is constructed on a Digital Equipment Cooperation'sVAX-11/780 minicomputer, running VAX/VMS operation system, at the Department ofElectrical Engineering, National Taiwan University. This implemetation is, infact, the extention of a successfully proven relational detabase systemaccomplished earlier at the same installation.

[科資中心編號]R72-TK7885-009
[題    名]
散佈資料庫系統的設計和建立(II)
Design and Implementation of a Distributed Database System. (II)
[研 究 者 ]
柯志昇
KE, JYH-SHENG
[機構名稱  ]中央研究院資訊科學研究所(ASII)
[經費來源  ]行政院國家科學委員會(NSC)
[補助金額  ]0
[補助編號  ]NSC71-0404-E011-02
[研究開始日期]1981/12
[研究結束日期]1982/11
[出版日期  ]1982/11
[建檔日期  ]1998/07
[原件提供  ]行政院國家科學委員會(NSC)
微縮片號: NSC71-0404-E011-02 3片 218幅
[科資分類號 ]IA
[關 鍵 詞 ]
協同控制;詢問處理;資料庫管理系統;分散式資料庫;區域網


Database management system;Concurrency control;Query processing;Distributed
database;Local area network

[摘    要]
本研究主要在建立一分散式資料庫系統並做相關之理論研
究。全部計畫工作共分下列四項: !!1.連結三架計算機並製作
必要之軟體程式,構成一區域網路。!!2.設計並製作一關係模
式資料庫管理系統之軟體程式。!!3.設計分散式資料庫系統
所必經之各種管理程式,包括協同控制和詢問處理。!!4.研究
協同控制,詢問處理,和通訊協定之理論。

The object of this project is to implement a distributed database system
andstudy the related theoretical backgrounds. The main research works of
thisproject are itemized as below:!!1. Design and implementation of
thecommunication software program for a heterogeneous local area computer network.!!2. Design and implementation of a relational database management system.!!3. Design and implementation of a distributed database system on top of thecomputer network as the relational database management system.!!4. Theoreticalstudy on the problems of concurrency control, query processing and communicationprotocol.

科資中心編號: RB9101-1117 
題名: 一致性檢查點之線上演算法之研究與設計 
The Study and Design of Efficient Online Algorithm for Identifying Consistent Checkpoints 
研究者: 陳隆彬 ; 吳毅成 
Chen, Loon-Been ; Wu, I-Chen 
機構名稱: 親民工商專科學校資訊管理科 
經費來源: 行政院國家科學委員會 
補助金額: 新台幣 267 仟元 
補助編號: NSC89-2213-E243-002 
研究日期: 2000/08 - 2001/07 
出版日期: 2001/09 
頁冊數: 4頁 
原件提供: 行政院國家科學委員會科學技術資料中心, NSC89-2213-E243-002 
科資分類號: IA0202;IA0301 
關鍵詞: 檢查點 ;分散式系統 ;分散式偵錯 ;容錯 ;線上演算法 
Checkpoint ;Distributed System ;Distributed Debugging ;Fault Tolerance ;Online Algorithm 
中文摘要: 分散式系統的除錯、測試、及容錯等問題通常藉由分析檢查點(Checkpoint)來解決。檢查點是一個處理機的執行狀態﹐儲存於高可靠性的媒體之中。當程式發生錯誤時﹐系統可從儲存媒體中載入檢查點後再繼續執行﹐以達到容錯的目的。在分散式系統中,一個"一致性全域檢查點"是能代表分散式程式的整體執行狀態的一組檢查點。若一個檢查點確定不屬於任一個一致性全域檢查點﹐則它是可刪除的。本計劃研究檢查點的online演算法﹐研究項目如下:( 1)設計online 演算法﹐一旦有新的檢查點產生﹐將所有的檢查點分為一致性或可刪除的。(2)如何 online 計算一致性全域檢查點。(3)研究檢查點演算法是否可以應用於分散式偵錯﹐以減少記憶體使用量並增快偵錯程式的執行速度。 
英文摘要: In a distributed system , analyzing checkpoints is essential for error recovery and debugging. A checkpoint is a recorded state of the execution of the program, stored in a reliable storage. Programs can be recovered or debugged by restoring the execution states saved in checkpoints. In a distributed system , a consistent global checkpoint is a set of checkpoints to represent the execution states of the distributed program. A checkpoint does not belong to any consistent global checkpoint is referred to as removable. In this project we investigate online checkpointing algorithms, research items include:(1) Design algorithms to classify the checkpoints as consistent or removable at each time when a new checkpoint is taken. (2) Design algorithms to calculate the set of consistent global checkpoints containing a specified checkpoint. (3) Investigate whether the checkpointing algorithm can be applied to identify useless debug information for distributed debugging. 


科資中心編號: RB9101-0947 
題名: 分散式系統查核點方法之最佳化研究 
A Study on Optimal Checkpointing in Distributed Systems 
研究者: 邱舉明 
Chiu, Ge-Ming 
機構名稱: 台灣科技大學電機工程系 
經費來源: 行政院國家科學委員會 
補助金額: 新台幣 484 仟元 
補助編號: NSC89-2213-E011-148 
研究日期: 2000/08 - 2001/07 
出版日期: 2001/10 
頁冊數: 6頁 
原件提供: 行政院國家科學委員會科學技術資料中心, NSC89-2213-E011-148 
科資分類號: IA0202;IA0301 
關鍵詞: 檢查點 ;最佳化 ;分散式系統 ;非因果關係基本間隔 ;返轉復原 ;無用查核點 
Checkpoint ;Optimization ;Distributed System ;Primary Noncausal Interval ;Rollback Recovery ;Useless Checkpoint 
中文摘要: 在分散式系統中,加入一組最少數目的強制查核點,以免除無用的查核點,不但可以避免資源的浪費,並使容錯技術更符合如即時性的需求。在本研究中,我們提出“非因果關係的基本間隔”(PNCI)的觀念,並証明只由PNCIs 中,就可找到一組最少數目的強制查核點,以使系統免除無用的查核點。我們進一步利用圖形理論,建立一套簡明的系統模式,藉由所建構的有向圖形,將“找尋一組關鍵強制查核點的問題”轉換成“解決經過特定點的路徑均不會形成迴圈的問題(SUBSET-FET)”。此外,我們也提出一套方法,以簡化原來問題的複雜度。經簡化後,在許多的情況下,可以有效率地找出其最佳解。 
英文摘要: Inserting a minimum number of forced checkpoints in a distributed system so as to eliminate useless checkpoints has been an important issue for the provision of fault tolerance capability. In this research, we first introduce the notion of primary non-causal intervals and show that these intervals are the only candidates that need be considered for inserting a minimum number of forced checkpoints. An efficient algorithm is then used to identify the primary non-causal intervals where forced checkpoints should be placed. Our algorithm first converts the original problem to another problem on a directed graph that may reflect the existence of useless checkpoints. The new problem can be efficiently solved using existing methods. Although our algorithm offers a near-optimal solution in general, optimal solutions could be obtained in many cases using reduction techniques. 

科資中心編號: RB9009-0185 
題名: 分散式即時環境上之時間驅動排程方法 
Time-Driven Scheduling Methodology on Distributed Real-Time Environment 
研究者: 薛智文 
Hsueh, Chih-Wen 
機構名稱: 中正大學資訊工程系 
經費來源: 行政院國家科學委員會 
補助金額: 新台幣 341 仟元 
補助編號: NSC89-2213-E194-038 
研究日期: 1999/11 - 2000/07 
出版日期: 2000/10 
頁冊數: 5頁 
原件提供: 行政院國家科學委員會科學技術資料中心, NSC89-2213-E194-038 
科資分類號: IA0301 
關鍵詞: 時間驅動排程 ;分散式系統 ;即時 ;端對端延遲 ;時距限制 ;時輪排程 
Time Driven Scheduling ;Distributed System ;Real Time ;End-Toend Delay ;Distance Constraint ;Pinwheel Scheduling 
中文摘要: 在分散式系統中常有針對端對端(end-to-end)的時間需求。其中對於最大端對端時間延遲的最小化是滿足服務品質(quality of service)最重要的時間限制(timing constraints)之一。樹狀結構(tree structure)是現行網際網路分散式系統中最常被使用的網路架構。在本研究中?A我們將分散式時輪排程(distributed pinwheel scheduling)中端對端最大時間延遲的最小化從在線狀結構(pipeline structure)上的應用擴展到樹狀結構上。另外,我們推導出兩個相鄰節點間最?j時間延遲的最小上限,並在線性的時間複雜度內算出兩個節點之間最大延遲的最小值。利用這兩項結果,我們就可以在分散式環境下更輕易且更快速地對即時的工作進行排程,使其符合時距限制(distance constraints)並且最小化其最大的時間延遲。 
英文摘要: Distributed real-time systems often have an end-to-end timing requirement. Minimizing the maximum end-to-end delay is one of the most important timing constraints we would like to guarantee for quality of services. The network topology of many distributed systems on Internet is often a tree structure. We extend our research on distributed pinwheel scheduling from pipeline structure to tree structure for minimizing the maximum end-to-end delay. We derive a tight maximum delay bound between two nodes and a linear-time algorithm to find the minimax delay between two nodes. With this bound and algorithm, it is easier and faster to schedule distributed real-time tasks with distance constraints and minimize the maximum end-to-end delay. 

科資中心編號: RB8910-0032 
題名: 在DCOM和CORBA環境下測試分散式軟體元件(II) 
Testing Distributed Component Objects on DCOM and CORBA Systems (II) 
研究者: 王建民 
Wang, Chien-Min 
機構名稱: 中央研究院資訊科學研究所(ASII) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 374 仟元 
補助編號: NSC88-2213-E001-024 
研究日期: 1998/08 - 1999/07 
原件提供: 行政院國家科學委員會科學技術資料中心 NSC88-2213-E001-024 
科資分類號: IA0202;IA0302 
關鍵詞: 分散式軟體元件 ;共通物件策託經紀人架構 ;元件測試 ;測試執行工具 ;分散式系統 ;物件導向程式設計 ;軟體元件 
Distributed Component Object ;Common Object Request Broker Architecture (Corba) ;Component Testing ;Test Execution Tool ;Distributed System ;Object Oriented Programming (Oop) ;Software Component 
中文摘要: 軟體元件技術已經逐漸成為軟體發展的主要方法,諸如Windows平台上的 COM和DCOM、Java平台上的JavaBeans和RMI以及分散式網路環境上的CORBA,都扮演著越來越重要的角色。本計畫延續先前對物件導向程式測試技術的研究,將發展一個針對DCOM和CORBA軟體元件的測試工具,它能夠針對不同的元件類型,自動產生適合的測試程式。所產生的測試程式能接受使用者的命令,動態地產生元件,對元件執行所選定的方法,使用者可以從而檢視執行之後的結果,以達到測試的目的。最後將利用此一測試工具來發展測試技術,以提高軟體元件的品質。 
英文摘要: Software component technology now has emerged as a main method for software development. COM and DCOM on the Windows platform, JavaBeans and RMI on the Java platform, and CORBA on the distributed environment are all playing more and more important roles. However, test techniques and tools for distributed software components are still not well studied. Thus, we will continue our research on object-oriented software testing and aim to develop a component-based test execution tool for distributed components. It can generate an appropriate test program automatically for various types of component objects. The generated test program, controlled by interactive users, allows users to create component instances at run time, execute interface methods on components, and display the contents of components. Hence, this tool can free test programmers from the burden of developing a large number of test programs. We will also develop methodologies for testing component objects. Combining both the test execution tool and the test methodology would yield an effective method for testing software components on distributed systems. 


科資中心編號: RB8902-0006 
題名: 分散式系統中有限返轉復原距離之查核點方法設計 
Checkpoint/Rollback-Recovery Scheme with Bounded Rollback Distance 
研究者: 邱舉明 
Chiu, Ge-Ming 
機構名稱: 台灣科技大學電機工程系(NTSTEET) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 455 仟元 
補助編號: NSC88-2213-E011-039 
研究日期: 1998/08 - 1999/07 
頁冊數: 6頁 
原件提供: 行政院國家科學委員會科學技術資料中心 NSC88-2213-E011-039 
科資分類號: IA0201 
關鍵詞: 分散式系統 ;檢查點間隔 ;重算回復 
Distributed System ;Checkpoint Interval ;Rollback Recovery 
中文摘要: 容錯能力對於大型分散式系統或長久運轉之計算是非常重要的課題。其中,查核點/返轉復原(Checkpoint/rollback)方法乃是可行的容錯技術。本研究提出一套能有效地調整查核點間隔之查核點/返轉復原方法,該法主要是利用計數值來觸發各個行程查核點的製作,計數值代表了查核間隔的次序、相依關係和返轉復原距離的資訊。我們的方法從幾個方面來考量適當地調整查核點的間隔,其一是限制送方的殘留間隔,使收方能維持一定的返轉復原距離;其二是在指定的間隔內限制製作查核點數目,使查核點的最大負荷受到限制;再者,當一行程收到訊息時,它能依訊息上所攜帶之計數值與其身上之計數值之大小,決定是否提前其下一查核點的製作。我們的方法,不僅使各行程滿足個別需求的最大返轉復原距離,並且每一行程僅需保留二個查核點,就可成功返轉復原。 
英文摘要: Fault tolerance is an important issue for large-scaled and long-running distributed systems. Checkpoint/rollback recovery has been an attractive approach for implementing fault tolerance for distributed systems. In this project, we propose a check point/recovery scheme that allows each process to achieve a bounded rollback distance during recovery by adjusting the checkpointing instants. A counter called control index is maintained by each node and is used to determine where the next checkpoint is to be placed. The adjustment of checkpoint instants is considered from several aspects. Firstly, we limit the residual interval at the sender of any message to no greater than I/sub s/, where I/sub s/ is the minimum of the regular checkpoint intervals of all of the processes. As a result, the induced rollback distance at the receiver can be bounded. Secondly, the number of checkpoints that can be taken within I/sub s/ is confined to limit the checkpointing overhead. When a process receives a message, it may determine whether to move its next checkpoint forward or not by comparing its control index with the value that is tagged to the message. Our scheme only requires each process to maintain at most two checkpoints for recovery. 


科資中心編號: RB8803-0054 
題名: 軟體元件技術研究---在DCOM和CORBA環境下測試分散式軟體元件(I) 
Testing Distributed Component Objects on DCOM and CORBA Systems (I) 
研究者: 王建民 
Wang, Chien-Min 
機構名稱: 中央研究院資訊科學研究所(ASII) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 312 仟元 
補助編號: NSC87-2213-E001-012 
研究日期: 1997/08 - 1998/07 
頁冊數: 4頁 
原件提供: 行政院國家科學委員會科學技術資料中心 NSC87-2213-E001-012 
科資分類號: IA0302;IA0301 
關鍵詞: 軟體元件 ;軟體測試 ;元件測試 ;分散式系統 ;測試執行工具 ;物件導向程式設計 
Software Component ;Software Testing ;Component Testing ;Distributed System ;Test Execution Tool ;Object Oriented Programming 
中文摘要: 軟體元件技術已經逐漸成為軟體發展的主要方法,諸如Windows平台上的 COM和DCOM、Java平台上的Java Beans和RMI以及分散式網路環境上的CORBA,都扮演著越來越重要的角色。本計畫延續先前對物件導向程式測試技術的研究,本年度的研究計畫將發展一個針對COM和DCOM軟體元件的測試工具,它能夠針對不同的元件類型,自動產生適合的測試程式。所產生的測試程式能接受使用者的命令,動態地產生元件,對元件執行所選定的方法(Method),使用者可以從而檢視執行之後的結果,以達到測試的目的。最後將利用此一測試工具來發展測試技術,以提高軟體元件的品質。 
英文摘要: Software component technology now has emerged as a main method for software development. COM and DCOM on the Windows platform, Java Beans and RMI on the Java platform, and CORBA on the distributed environment are all playing more and more important roles. However, test techniques and tools for distributed software components are still not well studied. Thus, we will continue our research on object-oriented software testing and aim to develop a component-based test execution tool for distributed components. It can generate an appropriate test program automatically for various types of component objects. The generated test program, controlled by interactive users, allows users to create components at run time, execute interface methods on components, and display the contents of components. Hence, this tool can free test programmers from the burden of developing a large number of test programs. We will also develop methodologies for testing component objects. Combining both the test execution tool and the test methodology would yield an effective method for testing software components on distributed systems. 


科資中心編號: RB8707-0018 
題名: 在分散式系統上設計一個對稱性之一致互斥演算法 
Design of a Symmetric Quorum Consensus Algorithm for Mutual Exclusion in Distributed Systems 
研究者: 張禾坤 
Chang, Her-Kun 
機構名稱: 長庚大學工商管理系(CGUGBA) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 125 仟元 
補助編號: NSC86-2621-E182-001-T 
研究日期: 1997/02 - 1997/09 
頁冊數: 3頁 
原件提供: 行政院國家科學委員會科學技術資料中心 NSC86-2621-E182-001-T 
科資分類號: IA0101 
關鍵詞: 分散式系統 ;分散式互斥 ;一致性演算法 
Distributed System ;Distributed Mutual Exclusion ;Quorum Consensus Algorithm 
中文摘要: 資源分享是分散式系統的主要優點之一,在分散式系統上的電腦可能會同時對某一共享資源發出使用的要求,此時就會有衝突產生,必須要有一個方法來解決這種衝突,此即所謂分散式互斥(Distributed mutual exclusion)的問題。一致性演算法(Quorum consensus algorithm)是解決分散式互斥所常用的方法,其通信的負擔可由Quorum size來估計。以前的演算法,不是只能針對特殊的結構,就是非對稱性(Asymmetric)的演算法,系統中的節點(電腦)負擔輕重不一,有些節點可能會造成瓶頸。本計畫已經設計出一個對稱性的演算法,而且不需特定的結構,得到一個兼顧效能與負載平衡的解決方案,進一步應用到分散式系統的設計上。 
英文摘要: Quorum consensus algorithms can be used to solve the problem of distributed mutual exclusion. The messages cost of a quorum consensus algorithm can be estimated by it quorum size, which is defined to be the average number of nodes in a quorum. Accordingly, an algorithm has less quorum size is preferred than that has a greater one. So far as we known, previous solutions are either restricted to special logical structures or asymmetric. An algorithm based on a specified logical structure also limits its application. On the other hand, an algorithm that is asymmetric may cause the most burdened node as a bottleneck. This project has designed a symmetric quorum consensus algorithm for distributed mutual exclusion. As a result of symmetric quorums, an efficient and load balanced solution for distributed mutual exclusion is achieved. 


科資中心編號: RB8705-0048 
題名: 容錯分散式計算系統(II) 
A Fault Tolerant Distributed Computing System 
研究者: 楊竹星 
Yang, Chu-Sing 
機構名稱: 中山大學資訊工程研究所(NCSUCIEG) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 783 仟元 
補助編號: NSC86-2213-E110-032 
研究日期: 1996/08 - 1997/07 
頁冊數: 7頁 
原件提供: 行政院國家科學委員會科學技術資料中心 NSC86-2213-E110-032 
科資分類號: IA0201 
關鍵詞: 容錯檔案系統 ;分散式系統 ;程序遷移 ;執行環境 
Fault Tolerant File System ;Distributed System ;Process Migration ;Execution Environment 
中文摘要: 由於微處理機的進步及網路技術的提升,使得計算環境從過去的單一大型主機演變成今天以網路連接數十個節點而成的分散式系統,具有資源分享、容錯、可擴充及可靠等特性。本計畫在Linux作業系統設計與實作負載平衡設施,其特色是提出Offset觀念為主的負載平衡策略。除了Queue length外,並將記憶體的使用情形及Disk I/O Queue length也納入工作負載指標(Load index)之一;定義Workload(工作負載)為三項指標加權相加,可動態調整其加權比重以適應不同的工作型態;並提出 Candidate group來平衡負載。根據負載平衡策略使用Initial job placement平衡負載;同時透過Process migration(行程遷移)動態調整負載,避免新工作同時匯入同一工作,造成瞬間負載過高,而且採用Forwarding(轉遞)的方法來解決遠端執行所引起的IPC問題。提高資源使用率、改善系統的效能的目的。 
英文摘要: In this project, we design and implement a load balancing facility in Linux kernel. Our feature is to propose a Candidate Group load balancing strategy based on the concept offset. Besides, the Queue Length, utilization of memory and Disk I/O Queue Length are considered as part of load indexes; Workload is defined as the sum of the three weighted load indexes, and could be dynamically adjusted to fit different job pattern. In the system architecture, every node's workload is evaluated and exchanged among nodes; whenever new jobs enter a node, our load balancing strategy is invoked and Initial Job Placement is used to balance load if necessary. At the same time, Process Migration is used to dynamically adjust workload to ease the situation where many jobs from busy nodes are placed at the same idle node so that the idle node is overloaded. Also, Forwarding is used to solve the IPC problems between jobs (or processes) caused by remote execution. To sum up, our goal is to reduce average job execution, increase system throughput so that the resource utilization is promoted, the resources distributed nodes are shared among nodes and the system performance is improved. According to the experiment result, our load balancing facility can balance workload among nodes effectively and improve system performance greatly. Compared with Threshold- based load balancing facilities(static and dynamic), ours has better performance and stability. 


科資中心編號: RB8707-0121 
題名: 分散式系統中負荷平衡化之研究 
A Study on Load Balancing in Distributed System 
研究者: 游坤明 
Yu, Kun-Ming 
機構名稱: 中華大學資訊管理系(CHUINA) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 383 仟元 
補助編號: NSC85-2213-E216-015 
研究日期: 1995/08 - 1996/07 
頁冊數: 83頁 
原件提供: 行政院國家科學委員會科學技術資料中心 NSC85-2213-E216-015 
科資分類號: IA0203;IA0201 
關鍵詞: 分散式系統 ;工作配置 ;平行虛擬機 ;搬移 ;負載平衡 ;類神經系統 
Distributed System ;Dispatch ;Parallel Virtual Machine (Pvm) ;Migration ;Load Balancing ;Neural System 
中文摘要: 在分散式系統中,多處理器電腦系統常被用來執行平行化的工作。在執行的過程中,我們必須考慮靜態的工作配置和動態的負載平衡兩個重要的問題,才能提昇系統的表現。在此研究計畫中,我們為PVM(Parallel virtual machine)這套分散式工具發展了一套利用人工智慧特性來解決上述問題的策略。在執行於PVM上應用程式的初始及執行期間,我們應用人工智慧的技巧來進行工作配置之最佳機器的預測,而後利用MPEP(McCulloch Pitts evaluation procedure)的觀念來進行動態的工作搬移以讓系統上各機器的負荷平衡化。為了應證此策略的成效,我們在實際的環境下進行模擬,同時我們也對此策略與傳統的三個方法在三個問題上所造成的影響作一比較。系統的評估分別在不同的工作量,行程到達率和計算機器數目三個不同實驗項目下進行。模擬結果顯示我們的策略比傳統的方法來的有效。 
英文摘要: Static task allocation problem and dynamic load balancing problem are important when we use the multi-processor computer system to execute parallelized jobs. In this research, we develop a new AI strategy to solve these two problems under the PVM (Parallel Virtual Machine). We use AI technique to predict the better allocation for PVM tasks in their initial assignment as well as running time. If some tasks executing on unsuitable hosts are detected, we use the MPEP (McCulloch Pitts Evaluation Procedure) to choose the suitable migration host. The measuring procedure proceed under homogeneous environment. We compare three problems' enhanced performance and overhead produced by our strategy with the other three conventional strategies. Our system is evaluated on the basis of the effect of the work size, process arrive rate and computing cluster size. The result of experiment show our strategy perform more effectively than the other chosen strategies. 


科資中心編號: RB8707-0006 
題名: 最大流量問題在分散式系統下之計算:推挽式前置流量方式
Distributed Algorithm for the Maximum Flow Problem: The Preflow Push- Pull Approach 
研究者: 李新林 
Lee, Sing-Ling 
機構名稱: 中正大學資訊工程系(CHCUCSIE) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 380 仟元 
補助編號: NSC85-2213-E194-005 
研究日期: 1995/08 - 1996/07 
頁冊數: 39頁 
原件提供: 行政院國家科學委員會科學技術資料中心 NSC85-2213-E194-005 
科資分類號: IA0101 
關鍵詞: 分散式系統 ;網路流量問題 ;最大流量 ;演算法 
Distributed System ;Network Flow Problem ;Maximal Flow ;Algorithm 
中文摘要: 最大流量問題是圖形演算法及作業研究中最關鍵的問題之一。目前解決此問題最有效率的演算法均衍生自Goldberg所發表之Preflow push演算法, Preflow push演算法的基本作法是將水流以漸進的方式由起點流往終點。針對單一流向之缺點,我們提出的Push-pull演算法包含了雙向正負Preflow的觀念。由於雙向的方式使得同時動作的節點增加,這使得程式平行度相對的提高。另外,由於正負Preflow的互相抵消作用使得程式執行量減少,在較少的時間內即可得出最大流量。另外,我們也討論了對於某一節點,水流自該節點流到終點之可能性。若某一節點之水流無法流到終點的話,則該節點可被凍結而不致於影響流往終點之最大流量。此種想法使得所有節點的高度值變化次數總和少於n/sup 2//2,進而降低計算成本。由於Push與Pull恰為二相對之運算,因此只需要小部分的修改,我們提出的Push-pull演算法可以很容易的應用於現有已知的Preflow push演算法。 
英文摘要: The class of the preflow push algorithms is the most efficient method to solve the maximum flow problem. In this thesis we introduce a bidirectional preflow algorithm to improve the parallelism of these algorithms. Instead of generating active vertices in single direction in Goldberg's algorithm, our algorithm generates active vertices on two reverse directions simultaneously. Since the number of active vertices will be increased during the execution, the parallelism of the algorithm can be improved. We also discuss the relation between distance and reachability of vertices. Using this observation, the total number of all labels changes marked on vertices is bounded in n/sup 2//2. Our algorithm can support all the class of preflow push algorithms developed so far such as FIFO, LIFO, highest distance, and maximal excess flow algorithms. 


科資中心編號: RB8705-0047 
題名: 分散式伺服器隨意視訊系統:子計畫(三)---隨意視訊通訊系統(I) 
Video-on-Demand Communication Systems 
研究者: 黃興燦 
Huang, Shing-Tsaan 
機構名稱: 清華大學資訊科學研究所(NTHUCOS) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 461 仟元 
補助編號: NSC85-2221-E007-019 
研究日期: 1995/08 - 1996/07 
頁冊數: 46頁 
原件提供: 行政院國家科學委員會科學技術資料中心 NSC85-2221-E007-019 
科資分類號: IA0202 
關鍵詞: 隨選視訊 ;通訊協定 ;分散式系統 ;通訊系統 ;視訊伺服器 ;互動式 
Video On Demand ;Protocol ;Distributed System ;Communication System ;Video Server ;Interactive 
中文摘要: 雖然隨意視訊已被研究了數年且有一些商業產品已經問世,不過它們之中不是因為使用特殊設計的硬體而使得成本太高,就是只利用現成的硬體,而在設計上並沒有對隨意視訊系統特性作考量,以致雖然成本不高卻無法提供滿意的服務。有鑑於此,雖然我們也使用現成的硬體設備,但在設計上我們將隨意視訊系統分成三個部分:即時系統核心、檔案管理系統及網路傳輸系統;並且考量隨意視訊的特性,針對我們的硬體設備特別設計這三個子系統使其能發揮系統的最大效能。本文將依次介紹我們隨意視訊系統的整個架構及資料流程,還有在系統中用來取代TCP/IP的網路傳輸協定,最後將討論互動式功能設計上的一些考量。本篇報告主要是由"Design and Implementation of A Video-onDemand Network Delivery System"這篇論文所構成,文中首先探討介紹整個隨意視訊的系統架構、系統資料流的過程、各子系統實作的細節及它們之間連結的關係。接著我們提出一個新的網路傳輸協定以取代TCP/IP。 
英文摘要: In the recent years, great advances in computers make numerous multimedia services possible and even more effective than before. Among these services, the topic VOD (Video-On-Demand) has been extensively proposed, researched and developed because of its great marketable value. In a VOD system, besides providing the video services, it also allows users to perform their desired actions individually, instantaneously and interactively. These actions can be named in terms of VCR functions: pause, resume, fast-forward and rewind. Due to extra overheads in existent systems, we separate our VOD system into three individual parts, including real-time kernel, file system and communication system and put careful considerations and special designs on them to fit the VOD requirements. In this thesis, we put much emphasis on network delivery system and we design a simple, light-weight transport protocol to utilize the network bandwidth by decreasing redundancy of existent protocols. The proposed protocol will emphasize two points, that is, reliability and flow control. Finally, the design considerations and implementation of interactive functions are included. 


科資中心編號: RA8510-2056 
題名: 一個在分散式系統上之可容錯物件應用環境
A Fault-Tolerant Object-Based Application Environment for Distributed Systems. 
研究者: 袁賢銘 ; 梁德容 
Yuan,Shyan-Ming ; Liang,De-Ron 
機構名稱: 交通大學資訊科學研究所(NCTUINSG) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 309 仟元 
補助編號: NSC84-2213-E009-093 
研究日期: 1995/02 - 1995/07 
出版日期: 1995/07 
頁冊數: 58頁 
原件提供: 行政院國家科學委員會科學技術資料中心 NSC84-2213-E009-093 
科資分類號: IA0301;IA0305 
關鍵詞: 分散式系統 ;容錯 ;物件導向程式設計 ;主從模式 
Distributed System ;Fault Tolerance ;Object Oriented Programming ;Client/Server Model ;Common Object Request Broker Architecture (Corba) 
中文摘要: 隨著分散式系統和物件導向技術的發展,以物件為基礎的分散式處理已廣受重視。因此,基於對一個標準化分散式物件計算環境的需求,包括Sun、HP等大公司在內的許多廠商組成了OMG(Object management group)、CORBA(Common object request broker architecture)正是它制定的第一個分散式物件交換的標準。提供了在分散式環境下,以物件導向模式發展Client-server應用程式的基礎。但是,這套標準並未包含容錯處理,若有一物件發生錯誤,就會影響到整個程式的進行。所以,Corba雖然提供較高階的分散式物件發展環境,但若要發展具Fault-tolerant能力的應用軟體,仍是一件繁重的工作,為了降低在Corba環境下撰寫具容錯能力的應用程式之成本,本研究計畫在研究在CORBA的環境下,如何利用Corba所提供的Mechanism,配合物件導向程式語言C++的特性(封裝、繼承、動態繫結),引進Checkpointing和 Recovery的技術,以提供Class library及Service object的方式,提供程式設計師更高階的發展環境,讓設計者可以用宣告一般C++物件的方法來宣告具有容錯能力的物件,不必操心繁複的容錯處理細節,卻又能兼顧程式的執行效率,而讓一般程式設計師能根據本身應用程式的需求,選擇適當的容錯物件類別來搭配,方便地設計出複雜的分散式容錯應用程式。Phoinix現已被實作在一個符合CORBA規格的實際產品Orbix平台上,藉由我們所提供的函式庫和改良版的Idl Compiler,設計者可以簡便地設計出具有容錯能力的分散式應用程式。 
英文摘要: With the development of distributed systems and object oriented technology, object-based distributed processing has been highly valued. Therefore, OMG (Object Management Group) is devoting itself to the standardization of distributed object exchange, and CORBA (Common Object Request Broket Architecture) is exactly its first standard for distributed object exchange. But the CORBA specification does not cover fault-tolerance. Consequently, if an object within some application occurs fault, the progress of the whole application will be obstructed. This report describes how to make use of the properties of the object oriented language C++ (Encapsulation, Inheritance, Dynamic binding) to embed checkpoints and audit trails technology into CORBA, providing programmer with higher level development environment- Phoinix, in which programmer can declare fault-tolerant object just like a conventional C++ object without worrying about how the fault- tolerance is implemented, and keep the application executing efficiently. Phoinix has been implemented on the platform Orbix, which is a full implementation of CORBA specification. Programmer can design fault-tolerant distributed application simply by means of the library and enhanced IDL compiler we provide. 


科資中心編號: RA8601-2045 
題名: 分散式共用記憶體及應用上之同步與一致性
Synchronization and Consistency in Distributed Shared Memory with Applications. 
研究者: 雷欽隆 
Lei,Chin-Laung 
機構名稱: 台灣大學電機工程系(NTUEELE) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 618 仟元 
補助編號: NSC84-2213-E002-007 
研究日期: 1994/08 - 1995/07 
出版日期: 1995/09 
頁冊數: 55頁 
原件提供: 行政院國家科學委員會科學技術資料中心 NSC84-2213-E002-007 
科資分類號: IA0301 
關鍵詞: 分散式系統 ;記憶體系統 ;一致性 ;同步 ;預取 
Distributed System ;Memory System ;Consistency ;Synchronization ;Prefetch 
中文摘要: 分散式系統因具備了透明化的資源共享,較佳的可靠度和可獲得性,以及高平行度與高效率的特性,而成為現代與未來機算機系統的主流之一。然而對程式設計師而言,在分散式系統中必需使用程式間之通訊與遠地程序呼叫,來統合各同做程序的運行,增加了程式的複雜度和設計者之負擔。但是如果系統能提供邏輯上的共享記憶體,則程式師可以利用共享記憶體來簡化設計方式。是故,將分散式共享記憶體建置於分散式系統中是非常必需的。在本計畫中,我們提出了分散式共享記憶體之理論基礎與應用技術的研究成果。我們會探討幾項課題,包括記憶體一致性定義模型與資料預取等。我們提出一廣義階層定理來處理各記憶體一致性定義模型,以及結合急進一致性定的資料預取。廣義階層定理主要是對不同的記憶體一致性協定,提供一連串的正規化定義,並建立一致性的階層架構。而在資料預取方面,我們提出結合急進一致性定的資料預取,利用一特別的同步運算元,使得被預取的共享資料,及啟動預取動作的最佳時刻,能被輕易辨識出來。 
英文摘要: Distributed systems are in the mainstream of modern and future computing systems because of their features such as transparent resource sharing, better reliability and availability, and higher parallelism and performance. On the other hand, programmers of distributed systems must use interprocess communication and/or remote procedure calls to coordinate concurrent processes, which increase both programming complexity and burden of programmers. However, if a logical scene of shared memory paradigm to simplify the design process. This is why building distributed shared memory into distributed systems is important. In this project, we present both theoretical bases and application techniques of distributed shared memory. We focus on several topics in this report, including memory consistency semantics models and data prefetching. And we provide a universal hierarchy theorem to address the distinct memory consistency semantics models, and a data prefetching approach combined with the aggressive consistence. The concept of universal hierarchy theorem is to present a series of canonical definitions of various memory consistency semantics and establish the consistency hierarchy. In the data prefetching, our approach is to support a special synchronization operation which the prefetchable shared data objects, as well as the best moment to launch a prefetching operation, can be easily identified. 


科資中心編號: RA8601-2042 
題名: 分散式即時計算環境的建立---總計畫 
Object Oriented Distributed Execution Environment for Real-Time System. 
研究者: 陳祈男 ; 侯廷偉 ; 黃悅民 
Chen,Chyi-Nan ; Hou,Ting-Wei ; Huang,Yueh-Min 
機構名稱: 成功大學工程科學研究所(NCKUENSG) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 1285 仟元 
補助編號: NSC84-2213-E006-116 
研究日期: 1994/08 - 1995/07 
出版日期: 1995/08 
頁冊數: 13頁 
原件提供: 行政院國家科學委員會科學技術資料中心 NSC84-2213-E006-116 
科資分類號: IA0301 
關鍵詞: 分散式系統 ;即時系統 ;物件導向 ;微核心 
Distributed System ;Real Time System ;Object Orientation ;Micro Kernel 
中文摘要: 本計畫的目標在於設計一個物件導向分散式即時執行環境(ORDERS)。整個ORDERS分為兩個子計畫:(1)分散式即時計算系統之研究。(2)再 OSF/DCE上物件導向發展環境之研究。在第一個子計畫中我們建立了一個物件導向分散式系統,從最底層的核心一直到上層的系統物件。而在第二個子系統中,則是針對執行環境以及系統物件的建構工具作一深入研究。 
英文摘要: The goal of this project is to design an Object-oRiented Distributed Execution environment for Real-time System (ORDERS). It consists of two sub-projects. The two sub-projects are: (1) The study on Distributed Real-time Computing System. (2) The Study on Object-oriented Software developing Environment under OSF/DCE. In the first sub-project, we establish an object-oriented distributed system including kernel and system objects. In the second sub- project, we make a complete research on the execution environment and the tools for establishing system objects. 



科資中心編號: RB8907-0033 
題名: 分散即時系統之偵錯、重組與回復 
Diagnosis, Reconfiguration and Recovery in Distributed Real-Time Systems 
研究者: 郭斯彥 ; 呂毓榮 ; 蔡智強 ; 黃立仁 ; 余文錦 ; 廖文彬 
Kuo, Sy-Yen ; Tsai, Ji-Chiang ; Hwang, Lih-Ren ; Huang, Li-Jen ; Liao, Wing P.N. ; Liao, Wen-Pin 
機構名稱: 台灣大學電機工程系(NTUEELE) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 380 仟元 
補助編號: NSC82-0408-E002-021 
研究日期: 1992/08 - 1993/07 
頁冊數: 45頁 
原件提供: 行政院國家科學委員會科學技術資料中心 NSC82-0408-E002-021 
科資分類號: IA0301 
關鍵詞: 分散式系統 ;多處理器系統 ;容錯系統 
Distributed System ;Multi-Processor System ;Fault-Tolerant System 
英文摘要: As we know, the technology of computer is progressing more and more. However, there is an obstacle as far as the electron's speed is concerned. As a consequence, the concept of "parallel" is introduced and quite a few multi-processor systems come about. Many problems also happen when it comes from single-processor systems to multi-processor systems. The more complex the system will become very low if the architecture is not well designed. Such problems slow the growing speed of multi-processor systems which are the current trend of the computer world. Communications in the distributed system are achieved by using the message-passing method. The hypercube architecture is the typical type for the message-passing system. So, the researches on hypercube multiprocessors are very important. The report concentrates mainly on how to achieve an fault-tolerant system because the property of fault-tolerance is very much desired. A fault-tolerant system has the following characteristics. Firstly, the reliability of the system surely become very high. Secondly, the system can provide "graceful degradation" in the occurrence of faults. Thirdly, this would allow the diagnosis subsystem to locate the faults without halting the whole system. Thus the precious system can be fully utilized without any waste. 


科資中心編號: RA8401-2043 
題名: 分散式系統下資訊安全環境之研究 
Information Security Environment for Distributed Computing Systems. 
研究者: 雷欽隆 
Lei,Chin-Laung 
機構名稱: 台灣大學電機工程系(NTUEELE) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 439 仟元 
補助編號: NSC82-0408-E002-019 
研究日期: 1992/08 - 1993/07 
出版日期: 1994/01 
頁冊數: 96 
科資分類號: IA;IA0300 
關鍵詞: 通行碼確認 ;存取控制 ;使用者階層 ;廣播網路 ;分散式系統 
Password Authentication ;Access Control ;User Hierarchy ;Broadcasting Network ;Distributed System 
中文摘要: 本計畫討論如何在分散式系統下發展一動 態的資訊安全環境。主要的研究包含三個課 題:控制策略(Control mechanism)、訊息傳輸策略 (Message transmission mechanism)及通行碼驗證策略(Password Authentication mechanism)。對於控制策略,首先我們提出三個有效率 的方法,分別在樹狀及階層式結構下配置給每 一個安全等級一把加密金匙。之後,我們提出 一個植基於算術編碼法的動態存取控制策略; 在我們的方法中,每一個使用者擁有一把使用 者金匙,而每一個檔案也被賦予一副檔案金匙 。接者,我們提出兩個用以決定使用者彼此階 層間關係的動態策略,其中之一是植基於整數 對的觀念,另一個策略則是加入了植基於背包 問題(Knapsack problem)的安全觀念。對於訊息傳輸策略,我們提出了一個保密 系統,它能確保訊息在一個能公開存取的分散 式系統中,安全地廣播訊息給合法接收者。我 們的方法中,假設每一位使用者擁有一對祕密 金匙;並且,由系統的管理者產生一個主要金匙 並交予發送訊息者,發送者經由它加密訊息並 傳送給合法的訊息接收者。最後我們討論通行碼驗證問題;首先我們 提出三種植基於幾何觀點的通行碼驗證策略--- 三角形、直線及單位圓;在我們提出的三個策 略中,使用了單程函數及美國數據保密標準(DES) 的觀念。之後,我們提出一種適用於遠程存取 的通行碼驗證策略,此法使用伶俐卡(Smart card) 並且不再需要於記憶體內儲存表格。 
英文摘要: In this project, we focused on the research and development of a secure dynamic computing environment for distributed systems. Accordingly, this dissertation concerns the studies of control mechanisms , message transmission mechanisms and password authentication mechanisms. For the control mechanisms, we propose efficient schemes for assigning cryptographic keys among a group of security classes organized in tree and hierarchy structures, respectively. Then a dynamic access control scheme based on arithmetic coding is proposed. In our scheme, each user is associated with a key called user key and each file is affiliated with a key called file key. Furthermore, we propose two dynamic user hierarchy schemes for determining relationships in information systems. First, a dynamic user hierarchy mechanism based on key pairs is proposed. And then, a secure dynamic user hierarchy mechanism is proposed. The main idea of it is inspired by the concept of trapdoor knapsack. For the message transmission mechanism, we propose a cryptosystem which can securely broadcast secret messages in a public access distributed system. It is assumed that each user is associated with a secret key pair. Besides, a master secret key is generated by a central authority server so that messages can be enciphered by the originator and sent to the legitimate recipients. Finally, the problem of password authentication is concerned. Three password authentication schemes based on triangles, straight lines and a unit circle encoding are proposed first. In our first three schemes, a one-way function and a strong cryptographic operation such as DES (data encryption standard) are adopted. Then, a remote access password authentication scheme using smart cards, which can verify the login password without verification tables, is proposed. The inspiration of our idea is derived from the Shamir's identity-based signature scheme and Elgamal's signature scheme. 


科資中心編號: RA8310-2053 
題名: 在OSI網路環境下設計一分散式偵錯環境(I) 
An OSI-Based Distributed Debugging System. (I) 
研究者: 劉安之 ; 楊東麟 
Liu,An-Chi ; Yang,Don-Lin 
機構名稱: 逢甲大學資訊工程系(FCUEIFE) 
經費來源: 交通部電信總局電信研究所(TCLG) 
補助金額: 新台幣 683 仟元 
補助編號: TL-82-310 
研究日期: 1992/08 - 1993/07 
出版日期: 1993/08 
頁冊數: 66 
科資分類號: IA;IA0301 
關鍵詞: 通訊協定 ;分散式系統 ;偵錯系統 
Communication Protocol ;Distributed System ;Debugging System 
中文摘要: 本計畫是將一分散式偵錯系統架構至OSI環 境中。此偵錯系統原本架構於TCP/IP之上。目的 是藉本計畫之設計與評估,一方面做為OSI之應 用實例,二方面對OSI與TCP/IP有深入的比較。我們的OSI環境是以ISODE(實驗性的ISO發展環 境)為主,輔以 SUNLINK-OSI軟體。系統偵錯時,再借 重網路分析儀SNIFFER的通訊協定分析。在轉換的過程中,我們儘量使用相同之使 用者介面(Socket),其參數之轉換採程式庫自動轉 換。如此可維持應用程式之透通性。成果中的另一項特色是分散式偵錯系統本 身即為一整合系統,包含使用者介面、資料庫 、網路管理等。以OSI方式製作,可以看出各模 組之介面需求。在實作上,使用者可選擇OSI或TCP/IP做為互 通的協定,提供了多通訊協定(Multi-protocol)的環 境。其效能評估將在後續計畫中提出。 
英文摘要: The project transforms a distributed debugging system into an OSI environment. The system was originally based on TCP/IP. Our goal is two-fold: one, use this system as an application subsystem on top of the OSI domain; two, compare and evaluate the two environments and the conversion between the two. Our OSI environment is built on ISODE, supported by the SUNLINK-OSI. We also make use of Sniffer for protocol analysis. To provide a transparent interface during system conversion, we use sockets in both domains, the parameter-passing of which are performed by a system library. Our first year result is an integrated system, consisting of a GUI, databases, a multiprotocol system, and network management. The system evaluation will be performed in the 2nd half of this project. 


科資中心編號: RA8310-2014 
題名: 分散式系統伺服元件設計與建構 
Design and Implementation of a Transaction Server for Distributed Systems. 
研究者: 袁賢銘 
Yuan,Shyan-Ming 
機構名稱: 交通大學資訊科學系(NCTSIFS) 
經費來源: 工業技術研究院(ITRI) 
補助金額: 新台幣 500 仟元 
補助編號: I82002 
研究日期: 1992/07 - 1993/06 
出版日期: 1993/06 
頁冊數: 40 
科資分類號: IA;IA0301 
關鍵詞: 異動伺服器 ;分散式系統 ;單一運算元 
Transaction Server ;Distributed System ;Atomic Operation 
中文摘要: 異動伺服器可以視為檔案伺服器的一種擴 充,它提供了單一運算元的能力。在處理共享 資源同時存取的問題上,我們使用Two-phase locking protocol去解決。而在分散式系統,一個異動可能會牽涉到多個不同的異動伺服器。因此 需要一個協調者去整合分散在不同異動伺服器 的動作。我們的系統使用Two-phase commit protocol 去委任分散式的異動。我們設計了 Timeout protocol和Restart protocol,使Two-phase commit protocol有 能力去處理機器故障的問題。 
英文摘要: A transaction server may be regarded as an extension or a variation of a file service that provides atomic operations on some or all of its files. The problem of the synchronizing accesses to shared objects while allowing a high degree of concurrence in our system will be resolved by using two-phase locking protocol. In distributed systems, one transaction may invoke more than one transaction servers that reside at different sites. Therefore, it needs to coordinate the networkwide commitment of all cohorts participating in a transaction. We implement our distributed transaction processing system which uses two-phase commit protocol to commit the distributed transactions. To extend the two-phase commit protocol to deal with failures, a timeout protocol and a restart protocol are also implemented in our system. 


科資中心編號: RA8507-2031 
題名: 分散式系統中重複行程容錯方法研究 
A Study on Replicated-Process Techniques in Distributed Computing Systems. 
研究者: 邱舉明 
Chiu,Ge-Ming 
機構名稱: 台灣工業技術學院電機工程技術系(NTITEET) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 245 仟元 
補助編號: NSC83-0408-E011-011 
研究日期: 1994/02 - 1995/07 
出版日期: 1995/07 
頁冊數: 45頁 
原件提供: 行政院國家科學委員會科學技術資料中心 NSC84-2213-E011-054 
科資分類號: IA0301 
關鍵詞: 容錯 ;分散式系統 ;重複行程 ;多重傳播 
Fault Tolerance ;Distributed System ;Replicated Process ;Multicast 
中文摘要: 在分散系統中,當部分的行程發生錯誤時,若無有效的容錯方法加以因應,往往會形成資源嚴重的浪費,而這種情形對於長久運轉的行程尤其明顯。重複行程的設計是提供分散式行程容錯能力的一個有效的方法。基本上,它是藉由行程的複製及合作來達到系統持續運轉的功能。在本研究中,我們提出一套有效的重複行程容錯方法,其目的除了能提供容錯能力外,更著重利用重複行程本來就有的多重性質,以達到提昇系統工作能力的目的,亦即讓分散式計算的完成速度得以增進。其中每一行程均複製成一行程組。每行程組中之各行程所收到訊之次序都是一樣,以保持計算之一致性。在此方法中,我們利用多重位址廣播(Multicasting)達成行程組與行程組間,以及每一組內行程間之聯繫,特別是所提出的容錯機制是自給自足的,亦即任一行程的錯誤均在其相關行程組內就得以補償,而不影響到其他行程組中之運作。所提出之錯誤處理法則很簡單又不會導致太大的耗損。尤其是,藉著行程重複特性,得以加速訊息處理次序的決定,也可以讓訊息的向外傳輸加快,以達到減短分散計算的時間。系統模擬的結果也應證了此法之可用性。 
英文摘要: This report presents a process-replication scheme which aims at providing fault tolerance as well as high performance to application tasks in a distributed computing system. A process is replicated to form a troupe. Identical delivering sequences for incoming messages are enforced on all replicas of a troupe to ensure consistency of the computing. Multicasts are used for inter- and intra-troupe communications. The fault-tolerant mechanism is self- contained in the sense that failures in a troupe are handled internally without relying on others for consistent recovery. Crash- handling procedure is simple with small overhead. In particular, the mechanism takes advantage of the redundancy of processes to expedite the completion of a distributed task by speeding up the determination of message delivering sequences and by letting the transmissions of application messages be carried out as soon as possible, at the expense of a small number of control messages. Simulations are carried out to evaluate the task completion time. 


科資中心編號: RA81-2169 
題名: 分散式即時計算系統之研究 
Study on the Distributed Real-Time Computing System. 
研究者: 陳祈男 ; 侯廷偉 ; 黃悅民 
Chen,Chyi-Nan ; Hou,Ting-Wei ; Huang,Yueh-Min 
機構名稱: 成功大學工程科學研究所(NCKUENSG) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 2001 仟元 
補助編號: NSC81-0408-E006-08 
研究日期: 1991/08 - 1992/07 
出版日期: 1992/09 
頁冊數: 112 
科資分類號: IA;IA03 
關鍵詞: 微核心 ;物件導向 ;即時 ;分散式系統 
Micro Kernel ;Object Oriented ;Real-Time ;Distributed System 
中文摘要: 本文五年計畫之第一年執行結果報告。本 計畫主要在探討分散式環境下,建立即時性之 作業系統。本年度完成之工作主要分為三部 份:(1)對各種有關之即時與物件導向作業系統 之評估。(2)一個新的即時性物件導向作業系統 發展模式之探討。(3)即時通訊部分所需要的之 硬體支援。在第一部分,一些較具代表性的即時作業 系統如ARTS、 HARTS、CHAOS、RTUNIX、Spring Kernel等分 別被提出來討論。而Clouds、Choices等物件導向 之系統亦被探討。這些評估,對我們設計的系 統是相當有幫助的。第二部分,則集中在結合物件導向之被動 模式(Passive model)、微核心、即時性等三者共容 的系統架構。這種架構至今仍是開放性問題 (Open problem),相當具有研究空間。最後一部分,則是點對點網路的即時性通 訊分析,並試製作一蟲蛀式尋徑器(Wormhole router) 用以了解其特性。 
英文摘要: This report describes the research of building a real-time operating system under distributed environments. The research is a five-year long term project. The achievements for the first year are categorized into three parts:(1) Survey of related real-time and object-oriented operating systems. (2) Building the developing model for a real-time object-oriented operating system. (3) Design of hardware support for real-time communication. In the first part, some typical real-time operating system are studied, such as ARTS, HARTS, CHAOS, RTUNIX, and Spring Kernel. Meanwhile, several object-oriented operating system, for example, Cloudes and Choices are also evaluated. Those studies are helpful to our model. The second part focuses on the system framework which integrates the concepts of object-oriented passive model, real-time, and micro kernel. The framework is believed as kind of open problem worthy to be studied. Finally, we study the point-to-point real-time communication and try to design a wormhole router to accommodate the communication. 

科資中心編號: RB9110-0005 
題名: 模糊派屈網路之專家系統建立及其在橋梁損傷評估之應用(II) 
Bridge Damage Assessment through a Fuzzy Petrinet Based Expert System(II) 
研究者: 蔣偉寧 
Chiang, Wei-Ling 
機構名稱: 中央大學土木工程研究所 
經費來源: 行政院國家科學委員會 
補助金額: 新台幣 273 仟元 
補助編號: NSC90-2211-E008-063 
研究日期: 2001/08 - 2002/07 
出版日期: 2002/10 
頁冊數: 6頁 
原件提供: 行政院國家科學委員會科學技術資料中心, NSC90-2211-E008-063 
科資分類號: IG0003 
關鍵詞: 模糊理論 ;派屈網路 ;橋梁損害評估 ;專家系統 ;不確定性 
Fuzzy Theory ;Petri Network ;Bridge Damage Assessment ;Expert System ;Uncertainty 
中文摘要: 本文提出一個新的模糊派翠網路與其演算法,並將其實作成混合式派翠網路系統-HPNS,以模擬我們為了同時處理專家系統內的不確定性與不精確性而提出的模糊法則式推理。此模糊法則式推理是一個融合可能性推理與模糊推理的綜合性推理方法。在本文的模糊派翠網路中,我們堅持原派翠網路的引發法則,那就是當一個Transition被引發之後,其所有輸入Place內的所有Token皆被移除,並有新的Token被適當的置於輸出Place內。此外,我們亦堅持原派翠網路內對 Marking的定義與語意。因此,原派翠網路已發展完善的分析方法,可檢視系統特性如Liveness、Safeness、Boundedness、Deadlock等,大多數皆可應用於本文的模糊派翠網路。本文提出的推理演算法不僅符合法則式推理的原則,原派翠網路的執行方式亦受到保持。 
英文摘要: There has been an increasing interest in developing bridge management systems (BMS) to alleviate the problems arising from old bridges. A key to success in BMS relies heavily on the reliability of the techniques adopted for damage assessment. It is widely recognized that the use of expert systems to capture the expertise and to mimic the reasoning patterns of experts on damage assessment is a promising direction towards BMS. However, several important issues need to be addressed in using expert systems for bridges damage assessment, such as: (1) linguistic assessment by experts usually takes the form of natural language that contains intrinsic imprecision, (2) uncertain and imprecise information involved in the damage assessment usually make the problem even harder, (3) the efficiency of fuzzy rule-based reasoning needs to be improved, and (4) an explanation facility is essential to increase the confidence about the assessment results. To address the issues discussed in the previous section, this research is aimed: (1) to develop a reasoning mechanism in an expert system, which is capable of dealing with uncertain and imprecise information. (2) to integrate expert systems with Petri nets. Petri nets with a powerful modeling and analysis ability are capable of providing a basis for variant purposes, such as knowledge representation, reasoning mechanism, knowledge acquisition, and knowledge verification. (3) to develop modularized rule bases for damage assessment of bridges and provide a fuzzy Petri net-based expert system tool in the Internet, based on the proposed methodology, that can assist the assessment of bridge damage. 


科資中心編號: RB9102-0776 
題名: 服務品質驅動式專家系統
QoS-Driven Expert Systems 
研究者: 何正信 
Ho, Cheng-Seen 
機構名稱: 台灣科技大學電子工程系 
經費來源: 行政院國家科學委員會 
補助金額: 新台幣 445 仟元 
補助編號: NSC89-2218-E011-015 
研究日期: 2000/08 - 2001/07 
出版日期: 2001/10 
頁冊數: 6頁 
原件提供: 行政院國家科學委員會科學技術資料中心, NSC89-2218-E011-015 
科資分類號: IA0407 
關鍵詞: 專家系統 ;服務品質 ;服務品質驅動式專家系統 ;推論引擎 
Expert System ;Quality Of Service (Qos) ;Qos Driven Expert System ;Inferencing Engine 
中文摘要: 傳統的專家系統在固定的輸入條件、執行環境以及處理程序下,大抵上只會產生一固定服務品質的輸出,此服務品質基本上是無法隨使用者或環境而改變的。為了增加專家系統的使用彈性及人性化程度,本研究旨在探討以服務品質來驅動專家系統的可能性。首先,由觀察人類專家及專家系統解題行為的差異,我們提出了三點與專家系統服務品質相關的因素,並以之為基礎,建構一可由服務品質驅動的雛型專家系統。實驗結果顯示,本雛型系統確可依使用者或環境的不同服務品質需求,調整其推理機構,達到運作更合理、更有效率、且更人性化的地步。 
英文摘要: A traditional expert system can only produce an output with a fixed level of quality, given identical input data, execution environment and the pre-defined inference procedure. In general, the output quality cannot be made respondent to the execution context. In order to have a reasonable output with flexible service quality from an expert system, we have compared the differences between how an actual expert solves a problem with that of an expert system and proposed three factors that affect the Quality-of-Service (QoS) of an expert system. Based upon that, we have developed a prototype expert system that can be driven by external QoS requests. Our experiments A traditional expert system can only produce an output with a fixed level of quality, given identical input data, execution environment and the pre-defined inference procedure. In general, the output quality cannot be made respondent to the execution context. In order to have a reasonable output with flexible service quality from an expert system, we have compared the differences between how an actual expert solves a problem with that of an expert system and proposed three factors that affect the Quality-of- Service (QoS) of an expert system. Based upon that, we have developed a prototype expert system that can be driven by external QoS requests. Our experiments show that the prototype system can automatically adjust its reasoning architecture to produce an output, the quality of which meets different QoS requests from either the user or environmental constraints. 


科資中心編號: RB8806-0072 
題名: 以Case-Base為主的模糊專家系統---以個人電腦維修為例
A Fuzzy Expert System through Case-Based Decision Aiding---Using Personal Computer Maintenance as an Example 
研究者: 徐松梅 
Hsu, Sung-May 
機構名稱: 台灣科技大學工業管理系(NTSTIMT) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 254 仟元 
補助編號: NSC88-2213-E011-026 
研究日期: 1998/08 - 1999/07 
頁冊數: 4頁 
原件提供: 行政院國家科學委員會科學技術資料中心 NSC88-2213-E011-026 
科資分類號: IA0407;IA0304 
關鍵詞: 模糊專家系統 ;經驗庫 ;規則庫 ;推論機 
Fuzzy Expert System ;Case Base ;Rule Base ;Inference Engine 
中文摘要: 本研究完成了一個以人工智慧中Case Base為主架構的模糊專家系統。本系統之主要功能如下: (1)經驗庫(Case Base)的管理與查詢:在經驗庫的管理上,使用了模糊數學中similarity關係的分類法將全部的Case分類成為不同類別的Case Base。使用者可依自訂.alpha.-cut的值來控制分類的粗細,進而影響搜尋速度的快慢。此種搜尋法,在速度上比一般專家系統在法則庫中做全部搜尋要快。在經驗庫的查詢上,先應用AHP法分析每個Case中各索引的重要性,將其轉換成權重,在建立索引間的模糊關係,進而解模糊關係方程式A。R=B以便在某一個case base中選取最佳的Case。(2)法則庫(Rule Base)的推論:運用了IF-THEN的Production Rule,使用模糊命題及似然推理(Approximation Reasoning)。(3)推理引擎(Inference Engine):對經驗庫中沒有收集到的問題,可在此時使用似然推理自行推論,求取答案。並可處理使用者回答「不知道」這個特定訊息。同時提出在使用法則庫推論時,當多個推論前提包含「不知道」這個特定訊息時,提出一個信心評比,可提供使用者參考,由使用自行決定是否採用此結論。最後,以電腦簡易維修為例,完成一個實際可行的模糊專家資訊系統。證實Case Base為主的專家系統構想是可以輔助人們解決實際的問題。 
英文摘要: This thesis demonstrates a fuzzy expert system which makes decision through case-based decision aiding. The fuzzy expert system consists of the following parts: (1) Management and query of case base: In the management part we use the similarity relation to cluster into different case groups. Users can control the number of case groups by adjusting the value .alpha.-cut. In the query part the AHP method is used to analyze the A。 R=B to get the best fitting case to solve the user's problems. (2) Rule Base: Use "IF-THEN" production rule to treat fuzzy proposition and approximate reasoning. (3) Inference Engine: For experiences with are not included in the case base groups, use approximate reasoning to do the inference. Finally an implement of this fuzzy system is given by using the maintenance of the personal computer as an example to show that the idea of case-based decision aiding in fuzzy expert system does work. 


科資中心編號: RB8803-0178 
題名: 網際網路智慧型助理之研究 
A Study on Intelligent Web Agents on the Internet 
研究者: 鍾興臺 
Chung, Hsing-Tai 
機構名稱: 淡江大學資訊工程系(TKUEIFE) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 439 仟元 
補助編號: NSC87-2213-E032-014 
研究日期: 1997/08 - 1998/07 
頁冊數: 3頁 
原件提供: 行政院國家科學委員會科學技術資料中心 NSC87-2213-E032-014 
科資分類號: IA0407;IA0202 
關鍵詞: 網際網路 ;智慧型代理人 ;全球資訊網 ;專家系統 ;分散式人工智慧 
Internet ;Intelligent Agent ;World Wide Web (Www) ;Expert System ;Distributed Artificial Intelligence 
中文摘要: 本研究的主要目的在針對目前全球資訊網路架構上,資訊擷取與整合等相關問題,提出改善解決的方法。為了達到這個目的,我們運用了智慧型助理(intelligent agents)的技術,並結合三層式架構,提出Intelligent Web Agents Model(IWAM),以延伸目前網際網路上運用最廣的主從架構。在IWAM中,除主從架構中的使用者與服務提供者外,另加入了中介者。而使用者、中介者、服務提供者分別有其個別的助理。助理間主要以KQML為溝通的介面。透過此架構,可以達成如下目標:一、使用者不必自己費時費力去蒐集資訊,只需透過其助理,便能取得有效的資訊。二、在服務提供者端,服務的介面除了一般的首頁、CGI介面之外,還可支援多種存取介面,提供較精確的資訊。三、中介者助理有資訊整合的能力,能將多個服務提供者助理送回的資訊加以過濾、整合。四、由於網站變動頻繁,中介者助理也能隨時監視或更新服務提供者的位址與其提供的服務。在實作上,著重於資訊整合部分,應用領域則在股市資訊上。 
英文摘要: The purpose of this thesis is to extend the current WWW architecture in support of better information retrieval and fuse. To archive this, we propose an Intelligent Web Agents Model (IWAM) which adopts the technology of "Intelligent Agents", and combine it with the concept of three-layer architecture, as an extension of the widely used client/server architecture. Our approach is to insert a broker between a client and a server. Each layer has its own agent, namely UserAgent (UA), BrokerAgent (BA), ServiceAgent (SA) respectively. Communications between agents are by use of the KQML. Under this architecture, the following goals can be achieved. First, the use of UAs reduces the load for users to collect information by themselves. Second, in addition to the homepage and CGI interface, a SA can support several interfaces for better and more precise information retrieval. Third, BAs have the capability of information filtering and fuse from different Web servers or SAs. Fourth, BAs can constantly monitor any changes Web servers or SAs might cause. In practice, our focus is on the information fuse, and the application domain is on the stock information. 


科資中心編號: RB8803-0163 
題名: 整合式網路管理專家系統
An Integrated Network Management with Expert System 
研究者: 劉安之 ; 楊東麟 
Liu, An-Chi ; Yang, Don-Lin 
機構名稱: 逢甲大學資訊工程系(FCUEIFE) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 1022 仟元 
補助編號: NSC87-2213-E035-002 
研究日期: 1997/08 - 1998/07 
頁冊數: 6頁 
原件提供: 行政院國家科學委員會科學技術資料中心 NSC87-2213-E035-002 
科資分類號: IA0407;IA0202 
關鍵詞: 專家系統 ;網路管理 ;警訊管理 ;錯誤管理系統 ;多網域網路 
Expert System ;Network Management ;Alarm Management ;Fault Management System ;Multi-Domain Network 
中文摘要: 我們使用簡單網路管理協定(SNMP)做為從事網路管理的標準,並分析網路資訊中的數值變化,利用簡單線性回歸函數,制訂出一套警訊指示函數(Alarm Indicator Function),描述SNMP變數受到網路不正常行為的影響時,該變化所代表的意義。利用這套函數,我們可以明確且輕易地訂定出管理警訊(Alarm),正可以解決大部分網管系統不明確的警訊定義。我們使用並改良MODEL(Managed Object DEfinition Language) [YEM196,YEM197]語言,來描述網路警訊與錯誤事件,並將警訊與網路拓樸緊密結合,透過編譯程式將其轉換成可以讓網管系統接受的管理資訊,包含警訊設定與管理物件組態。透過我們的方法,目前已可成功判別出十種網路錯誤,同時亦提供實驗數據與分析過程,方便使用者參考與瞭解本系統的實際運作;另外,本計畫亦透過視覺化介面,展示目前網路上發生的錯誤警訊與網路拓樸的關係。 
英文摘要: The alarm management plays an important role in managing a network effectively. To administer an operational network, one focuses on network operations. Today, lots of the network activities are on TCP(UDP)/IP. Hence we need to manage the network by monitoring the corresponding TCP(UDP)/IP MIBs. Especially when we can analyze the network alarms according to the fluctuation of TCP(UDP)/IP MIBs. We focus on the value change of each monitored MIB at consecutive intervals, which is denoted by .DELTA.MIB. In this project we propose a method to analyze the network alarms by monitoring .DELTA.MIBs. We devise the relations among network alarms in order to provide information for alarm correlation. Most of the network management do not emphasis the ability for correlating network alarms. They do not properly organize and define the formats of network alarms. By defining the network alarms in a statistical model, we can analyze the behavior of the network infrastructure. Hence we can figure out the relationships between related network events more easily. Based on the analysis, we develop an integrated alarm-processing module called alarm transformation system. It is used to provide automatic network fault management in the near future. The alarm transformation system transforms the predefined network alarms, which specify the event-triggering conditions according to the analytical results, into the event settings of network management systems. By observing the captured events in the network management system, we are able to figure out the possible network anomalies. Our system can provide not only the traditional management functions, but also the advanced features for managing network alarms. In our lab environment, we can successfully recognize ten network faults of TCP(UDP)/IP. We can also find out the event relations in each single fault. Meanwhile, we combine the topology information with alarm settings. The most important feature of our method is the simplicity of configuring and analyzing the events. The challenge ahead is the generality under an open and heterogeneous environment. In order to adapt the general network environment, we still have to improve the analytical model. 


[科資中心編號]RB8707-0348
[題    名]
專家系統知識庫知識驗證的邏輯檢驗工具
A Comprehensive Logic-Checker in Knowledge Verification/Validation
for Rule-Based Expert Systems
[研 究 者 ]
李錫智
Lee, Shie-Jue
[機構名稱  ]中山大學電機工程系(NCSEELE)
[經費來源  ]行政院國家科學委員會(NSC)
[補助金額  ]424200
[補助編號  ]NSC85-2213-E110-037
[研究開始日期]1995/08
[研究結束日期]1996/07
[頁 冊 數 ]122頁
[語    文]英文
[建檔日期  ]1999/10
[出版情形  ]完成日期:民國85年7月
[原件提供  ]行政院國家科學委員會科學技術資料中心
NSC85-2213-E110-037
[科資分類號 ]IA0407
[關 鍵 詞 ]
知識驗證;法則基專家系統;知識正確性分析;規格;Petri網理論;邏輯推演;定理
證明;可靠度

Knowledge validation;Rule based expert system;Knowledge verification;
Specification;Petri-net theory;Logic deduction;Theorem proving;
Reliability

[摘    要]
由於專家系統在諸多領域均有卓越的表現,使得專家系統的使用愈來愈普遍,也愈來愈重要。知識庫的正確性,必須有賴自動化的知識庫檢驗工具來達成。專家系統知識庫的檢驗包括可用性檢驗/規格驗證兩部分。本計畫是從邏輯的觀點來進行知識庫的可用性檢驗/規格驗證兩件工作,藉由邏輯的子句將可用性檢驗/規格驗證與知識庫的錯誤形態予以正規化,再利用自動化推理的技術來驗證正規化後的子句。並以某一特殊的圖形理論---Petri Net來輔助邏輯在推演時效率的不足。最後利用這些方法發展出一個完整的工具來進行法則式知識庫的可用性檢驗/規格驗證。

Due to its success in many practical applications, expert
systems are more and more popular. Reliability of knowledge bases
must be achieved automatically and efficiently. Investigating the
reliability of an expert system has to perform the tasks of knowledge
validation and knowledge verification. This project is to investigate
both knowledge validation and knowledge verification of rule-based
expert systems by means of formal logic deduction. We plan to
formalize the types of improper knowledge, the specification from
users, and the tasks of knowledge validation/verification by logic
sentences and to prove these logic sentences formally in an automated
reasoning system. Moreover, based on the Petri-Net theory, we also
plan to develop a control mechanism which guides the automated
reasoning when searching a proof. Finally, a comprehensive tool which
performs completely knowledge validation and verification for rule-
based expert systems is designed.

[科資中心編號]RA8510-2141
[題    名]
專家系統與類神經網路之整合及其應用
The Integration of Expert Systems and Neural Networks and Its
Applications.
[研 究 者 ]
蘇木春
SU,MU-CHUN
[機構名稱  ]淡江大學機械工程系(TCUEMEE)
[經費來源  ]行政院國家科學委員會(NSC)
[補助金額  ]226000
[補助編號  ]NSC83-0408-E032-006
[研究開始日期]1994/02
[研究結束日期]1995/07
[出版日期  ]1995/07
[頁 冊 數 ]58頁
[語    文]中文
[建檔日期  ]1997/05
[出版情形  ]研究報告,58頁,民國84年7月
Report, 58p., Jul. 1995
[原件提供  ]行政院國家科學委員會科學技術資料中心
NSC84-2213-E032-005
[科資分類號 ]IA0407
[關 鍵 詞 ]
專家系統;神經網路;機器學習;模式識別

Expert system;Neural network;Machine learning;Pattern recognition

[摘    要]
在人工智慧的發展過程中,專家系統無疑
地是其中最具實用價值的研究領域之一。目前
專家系統已被廣泛地應用於醫學診斷、控制系
統等實際問題上,然而,於建立專家系統的過程
中,通常都會遇到一項極大的困難---如何從人
類專家中獲得所需之專家知識(Expertise)及如何
將此知識表示成專家規則(Production rules)。
近幾年來,類神經網路廣泛地引起學者們
的注意,因為類神經網路具有許多吸引人的特
性---強韌性(Robustness)、廣義性(Generalization)、
錯誤容忍性(Error tolerance)及範例學習(Learn by
examples)等。然而,它的最大的缺點就是它只給
予使用者答案而無法提供邏輯上的解釋,這個
缺點於許多實際運用上是不容忽視的,例如在
醫學診斷上,對如何能得到診斷的解釋是極有
參考價值的。
基於對專家系統與類神經網路的詳細分析
與評估後,本計畫將研討如何將這兩種各具特
色的系統整合成具有兩者優點的網路型專家系
統(Expert networks)。在此計畫中我們將提出嶄新
的類神經網路之架構及其學習演繹法則,以學
習過後之類神經網路能立即地提供建立傳統專
家系統所需的專家規則,此種新系統將能有效
地應用於醫學診療系統、模式識別(Pattern
recognition)、估測(Estimation)等問題上。

Rule-based expert systems are rather practically developed in
the artificial intelligence (AI) field. They have been greatly
applied in problems such as medical diagnosis systems and control
systems etc. However, a major difficulty in the design of expert
systems is the process of acquiring the requisite knowledge in the
form of production rules.
Recently, neural networks are attracting a lot of interest in
the scientific community because of their dynamic nature: robustness,
capability of generalization, and fault tolerance. Nevertheless,
there is a great disadvantage in neural networks they do not justify
their responses on the basis of explicit logic rules. This feature
is particularly important in medicine.
Based on the evaluations and discussions of both expert
systems and neural networks, this project is focused on how to
integrate these two different approaches. In this project, we will
develop some new neural network architectures and corresponding
learning algorithms. From these trained neural networks, we can
easily and efficiently build traditional expert systems by use of
the production rules extracted from the trained neural networks.
This new approach will be efficiently applied in medical diagnosis
systems, pattern recognition, and estimation problems.


[科資中心編號]RA8510-2136
[題    名]
以隨機模擬推理法設計一專家系統
A Design of an Expert System by Stochastic Simulation Reasoning
Method.
[研 究 者 ]
黎漢林
LI,HAN-LIN
[機構名稱  ]交通大學資訊管理研究所(NCTUIFMI)
[經費來源  ]行政院國家科學委員會(NSC)
[補助金額  ]204000
[補助編號  ]NSC83-0408-E009-024
[研究開始日期]1993/08
[研究結束日期]1994/07
[出版日期  ]1994/07
[頁 冊 數 ]93頁
[語    文]中文;英文
[建檔日期  ]1997/05
[出版情形  ]研究報告,93頁,民國83年7月
Report, 93p., Jul. 1994
[原件提供  ]行政院國家科學委員會科學技術資料中心
NSC83-0408-E009-024
[科資分類號 ]IA0407
[關 鍵 詞 ]
專家系統;人工智慧;隨機模擬;推理

Expert system;Artificial intelligence;Stochastic simulation;Reasoning

[摘    要]
本研究利用推理與學習技術發展一隨機模
擬法,以處理多重結果之因果網路下的診斷問
題。基於此一隨機模擬推理新法,本研究設計
一雙向推理專家系統,其功能包括:(1)規則學習:
由事例中導出規則。(2)正向推導:由已知(或狀
態未明)的原因推導出可能的症狀。(3)反向推
導:由已知(或狀態未明)的症狀反推回可能的原
因。本研究發展之專家系統較目前常用者更適
於處理不確定性下之大型因果網路下的診斷問
題。

This study develops a stochastic simulation method, based on
reasoning and learning techniques. The method can solve the
diagnosis problems of multiple symptoms and multiple causes. A
bidirectional expert system is then proposed using the stochastic
simulation method. The functions of such an expert system contain (1)
learning from examples, (2) deductive reasoning, (3) abductive
reasoning. The proposed expert system is more convenient than
current expert systems for solving large-scale diagnosis problems.


[科資中心編號]RC81-1499
[題    名]
關連式資料庫查詢最佳化學系統與專家系統的 整合
Integrate Expert System into Query Optimization Subsystem of Relational
Database.
[研 究 者 ]
林東清
LIN,TUNG-CHING
[機構名稱  ]中山大學資訊管理研究所(NSYNIFMG)
[經費來源  ]行政院國家科學委員會(NSC)
[補助金額  ]337000
[補助編號  ]NSC81-0301-H110-14
[研究開始日期]1991/08
[研究結束日期]1992/07
[出版日期  ]1992/07
[語    文]中文
[建檔日期  ]1994/05
[出版情形  ]

[科資分類號 ]SM0404
[關 鍵 詞 ]
專家系統;關聯式資料庫;查詢最佳化子系統

Expert system;Relational database;Query optimization

[摘    要]
自70年代以後,由於其具有整合性、分享性 等優點,資料
庫系統逐漸取代了傳統的檔案處 理系統,而在三種資料庫中(
階層式、網路式、 關聯式),由於其具有資料最獨立性、高階
的使 用者介面、最好的彈性,關聯式資料庫已確定 成為資料
庫的主流。
關聯式資料庫的有效率與否,其中最主要 牽涉到的便是
查詢最佳化子系統(Query Optimization Subsystem)。同樣的一個查詢
,不同的資 料擷取路徑,其所費的時間可能相差數十倍。
傳統的查詢最佳化處理,一直局限在實體 面的評估與計
算,而未注意到邏輯面資訊的可 資利用。本計畫即欲整合專
家系統與查詢最佳 化子系統,其主要精神在於利用邏輯層面
的資 訊,由專家系統轉化使用者原來輸入的查詢 (Query),而使
其成為一個新的、更有效率而時間 最短的查詢,其結果能大
大的節省查詢時間(甚 至到達90%以上)提高資料庫的使用效率


Since 1970's, Data Processing technique have evolved into a new era, and
Database Systems, with its advantages on integration, sharing, data
independence, and so forth, have gradually, replaced the conventional file
processing systems. Among these three kind of Database Systems, which are: Hierarchy, Network and Relational Database. Relational Database is valued as the best one for its simplicity, flexibility, formality and high-leveled user interface, and therefore becomes the mainstream of Database Systems. The efficiency of Relational Database primarily depends on its Query Optimization Subsystem. For each query, the cost of fetching an information can vary widely, depending on how the design of Query Optimization Subsystem is. Conventional Query Optimization techniques are limited within two aspects: First, the information used to evaluate the query comes only from physical level. Second, only one query representation form exists. The major motivation of this project is trying to release the limitation of conventional techniques mentioned previously. In doing so, an Expert System will be created and embedded into the Query Optimization Subsystem. With theuse of higher (i.e. logical) rather than physical-leveled information, the Expert System will be able to transform the original query into a more efficient, more cost- effective form, whence the information accessing cost ofthe Database will reduce dramatically.


[科資中心編號]RA78-2066
[題    名]
專家系統之知識庫設計
Knowledge Base System Design for Expert Systems.
[研 究 者 ]
郭德盛;劉寶鈞
KUO,TE-SON; LIU,BAW-JHIUNE
[機構名稱  ]台灣大學電機工程系(NTUEELE)
[經費來源  ]交通部電信總局電信研究所(TCLG)
[補助金額  ]538800
[補助編號  ]無
[研究開始日期]1988/02
[研究結束日期]1989/01
[出版日期  ]1989/01
[頁 冊 數 ]100
[語    文]中文
[建檔日期  ]1998/07
[科資分類號 ]IA;IA0407
[關 鍵 詞 ]
標的式模型;知識庫系統;專家系統

Object-oriented model;Knowledge-based system;Expert system

[摘    要]
本計畫旨在利用標的式方法模塑一知識庫系統,標的式方
法將系統看成是由一標的所構成.標的可依其特性而分別
為某個類別之成員,而類別又可依其需要加以擴充,將擁有更
多相同性質之標的,再集合成一個次類別.同時,使用者只能藉
由訊息之傳遞,透過類別所定義的運算來存取及處理該類別
之成員.在模塑方法中,我們把知識庫系統看成是由相互作用之
Knowledge object所構成的,而每一Knowledge object分別由behavior part,knowledge base part和monitor part所構成,behavior part用來描述object之行為,它包含類似於Smalltalk之methods;Knowledge base part則包含類似於Prolog的fact及rule;monitor part則monitor和控制object,我們並探討了這三個part彼此間的關係.

A method for an object-oriented modeling of knowledge-based systems is
proposed. In this modeling method, a knowledge-based system consists of
cooperative knowledge objects, where each knowledge object consists of a behavior part, a knowledge base part and a monitor part. The behavior part ofan object contains methods like those in smalltalk; the knowledge base part contains rules and facts like those in Prolog; and the monitor part monitors and controls the object. The relation between class and object, the relation between the behavior part and knowledge base part are described.

[科資中心編號]RB8705-0082
[題    名]
分散式象棋專家知識擷取系統之研製
Design and Implementation of a Distributed Knowledge Acquisition
System for Chinese Chess
[研 究 者 ]
許舜欽
Hsu, Shun-Chin
[機構名稱  ]台灣大學資訊工程系(NTUECSIE)
[經費來源  ]行政院國家科學委員會(NSC)
[補助金額  ]252000
[補助編號  ]NSC85-2213-E002-060
[研究開始日期]1995/08
[研究結束日期]1996/07
[頁 冊 數 ]70頁
[語    文]中文
[出版情形  ]完成日期:民國85年10月
[原件提供  ]行政院國家科學委員會科學技術資料中心
NSC85-2213-E002-060
[科資分類號 ]IA0304
[關 鍵 詞 ]
象棋;知識庫;分散式知識擷取;專家系統

Chinese chess;Knowledge base;Distributed knowledge acquisition;Expert
system

[摘    要]
如何借重象棋專家知識,提升電腦象棋程式的棋力,一直是我們研究的方向之一。至目前為止,已研製完成電腦象棋知識庫的雛形,並且應用在個人電腦查詢棋譜上。但由於資料來源,局限於棋書上的棋譜以及專家棋士的對局,始終無法迅速的擴充知識庫的規模,以提供電腦象棋程式有效的運用。為更廣泛的吸收全國各地象棋專家的知識,本計畫以主從架構的模式,在網際網路上設計製作一擷取系統。系統中包含數個子系統,其中網路對戰子系統,提供網路對奕及棋局資料收集的功能。而棋譜資料庫子系統,具有協助初學者學習象棋及專家輸入新棋步的功能。同時,我們亦針對原有知識庫,強化其內部結構,以期能容納更多的知識。

One of the objectives of our research has always been to find
ways to benefit from the knowledge and skill of Chinese chess experts
so as to raise the playing skill of the Chinese chess computer
programs. We have now made a good start in completing a Chinese chess
knowledge-base, and have applied it to personal computers for Chinese
chess book database. However, since the present source of data is
limited to the plays, strategies and famous matches played by chess
masters that have been published in the chess reference books, the
speed of building up such a knowledge-base that is effectively useful
to a Chinese chess computer program has been slow.
This project aims at designing and implementing an acquisition
system on the Internet with the client/server architecture so as to
absorb more broadly the knowledge of Chinese chess masters and
experts nationwide. This system includes several subsystems, of which
the subsystem for playing among the networks will provide the
functions of matches between networks and accumulation of data of
matches. The Chinese chess book database subsystem will have the
function of helping beginners to learn the game and experts to input
new moves. At the same time, we will bolster the internal structure
of the existing knowledge-base so as to be able to absorb more
knowledge.

[科資中心編號]RC8607-1258
[題    名]
專家系統及法則歸納技術運用於智慧型選股系統之研究
An Intelligent Stock Selection System Based on the Expert System and
Rule Induction Technologies.
[研 究 者 ]
楊朝成;林煜宗
YANG,CHAU-CHEN;LIN,YU-TSUNG
[機構名稱  ]台灣大學財務金融系(NTUNFIN)
[經費來源  ]行政院國家科學委員會(NSC)
[補助金額  ]283500
[補助編號  ]NSC85-2417-H002-036-E7
[研究開始日期]1995/08
[研究結束日期]1996/07
[頁 冊 數 ]23頁
[語    文]英文
[建檔日期  ]1998/03
[出版情形  ]完成日期:民國85年10月
[原件提供  ]行政院國家科學委員會科學技術資料中心
NSC85-2417-H002-036-E7
[科資分類號 ]SG0306;SM0404
[關 鍵 詞 ]
股票選擇;人工智慧;模糊專家系統;類神經網路;法則歸納

Stock selection;Artificial intelligence;Fuzzy expert system;
Artificial neural network;Rule induction

[摘    要]
人工智慧(Artificial Intelligence, AI)在今日已經很廣泛的被應用在
各個領域上。在財務金融方面的,由於要考量的變數很多,更需要人工智慧的技術來輔助。專家系統(Expert System)和類神經網路(Neural Network)是目前人工智慧中最大的兩個主流,然而現有的研究中,大都是使用單一技術加以改良,這兩種方法,各有其優點:專家系統在知識表達及解釋上的能力很強;類神經網路則擁有強大的學習能力。有效的結合專家系統、類神經網路,將可以改善原有之缺點。本研究計畫將模糊專家系統、類神經網路以及機器學習(Machine Learning)做一個整合,使其互補彼此之缺點,並且還要將整個問題以兩個步驟來完成:中、長期趨勢預測部分和短期趨勢買賣預測。第一個部分以模糊專家系統為主體,加以類神經網路來輔助。系統內的初始法則(initial rules)將取自書本和專家的意見,再利用類神經網路的學習功能隨時或定時的對這些法則進行修正;第二個部分以法則歸納(rule induction)為主體,並加上個例基礎推理(case-based reasoning)的概念而完成。本報告將這套智慧型的證券投資支援系統,實際驗證於台灣股票巿場,以1992到1994為實證期間,平均而言,其投資效益比"buy and hold"方式高。

This paper presents an intelligent stock trading decision
support system that is capable of forecasting buy and sell signals
according to the long-term up and down trends of the market. The
system is aimed at providing an investor with long-term portfolio
construction strategies based on the fuzzified daily stock
information including the opening, highest, lowest, and closing
prices and the trading volume that are turned into numerous technical
indices as input to the system.
There are two major components in this system. A fuzzy expert
system, which is built on empirical stock trading rules and is
responsible for identifying the long-term up and down trends of
stocks. The second component is composed of a collection of rule
induction trees, of which each is built upon the past data patterns
of some particular stock and is capable of producing a short-term buy
or sell signal for that stock.
The performance of the system is evaluated using the Taiwan
stock market information of the past three years. In the testing
period of 1992 to 1994, the annual rates of returns generated by the
proposed system were higher than those using the simple buy-and-hold
strategy. The performance of the proposed system is also superior to
the major closed-end mutual funds in Taiwan.

[科資中心編號]RB8907-0067
[題    名]
超媒體專家決策支援系統在軟體專家管理領域之應用研究
The Development and Application of a Hypermedia-Based Expert
[研 究 者 ]
余千智; 顧綱基; 陳明星; 武美旻
Yu, Chien-Chih; Ku, Kang-Chi; Chen, Ming-Hsing; Wu, Mei-Min
[機構名稱  ]政治大學資訊管理系(NCCCMIS)
[經費來源  ]行政院國家科學委員會(NSC)
[補助金額  ]263000
[補助編號  ]NSC82-0415-E004-005
[研究開始日期]1993/02
[研究結束日期]1994/01
[頁 冊 數 ]102頁
[語    文]中文; 英文
[建檔日期  ]2001/02
[出版情形  ]完成日期:民國83年3月
[原件提供  ]行政院國家科學委員會科學技術資料中心
NSC82-0415-E004-005
[科資分類號 ]IB0002; IB0402
[關 鍵 詞 ]
專案管理; 決策支援系統; 超媒體; 軟體發展; 專家系統

Project management; Decision support system; Hypermedia; Software
development; Expert system

[摘    要]
軟體專案管理為軟體開發專案成敗的關鍵,軟體專案管理流程之相關決策及工作項目,包含確認軟體發展目標,制訂軟體需求規格,建立軟體工作分解結構,成立軟體專案組織,估計軟體開發成本及風險,排定軟體發展工作時程,配置軟體專案資源,控制專案進度與品,製作系統文件,差異分析及問題診斷處理等,需要整合軟體專案中與品質、工作、成本、人力、時程、工具等有關的各種媒體資料、決策模式與專業知識,同時建立一個具有圖像化交談式介面的親和性軟體發展及專案管理系統環境,才能使整體軟體專案管理流程的推展兼有效率與效益。本研究的目的即在整合物件導向超媒體與專家決策支援系統等方法,用以探討超媒體軟體專案管理專家決策支援系統環境的建構與應用。一個採用軟體整合方法所建立的軟體發展成本估計與風險評估原型系統,將用來做為本研究的可行性驗證。

Software project management is the key to the success of
software development projects. Decisions and tasks in the software
project management process include identifying software development
objectives, specifying software requirements, designing software
development work breakdown structure, organizing project team,
estimating software development costs and risks, allocating project
resources, scheduling software development works, controlling
project qualities, documenting software development process, and
diagnosing as well as solving project problems, ... etc. Carrying
out the complete management and decision process efficiently and
effectively need to integrate data, models, and knowledge about the
project qualities, works, costs, labors, schedules, and tools, and
to provide an user friendly graphical interactive software
development and project management system environment. The goal of
this research is to integrate object-oriented method, hypermedia
technology, as well as expert decision support system approaches
for the development and application of an intelligent software
project management system environment. A prototyped software
development cost estimation and risk evaluation system is also
provided to validate the feasibilities of this research.

[科資中心編號]RB8904-0028
[題    名]
專家系統技術在電腦網路設計、規劃上的實作
Implementation of an Expert System in Computer Network Design
[研 究 者 ]
李錫智; 周宏昇; 余政哲; 鐘木騰
Lee, Shie-Jue; Chou, Hung-Sheng; Yu, Cheng-Che; Chung, Mu-Teng
[機構名稱  ]中山大學電機工程系(NCSEELE)
[經費來源  ]行政院國家科學委員會(NSC)
[補助金額  ]474000
[補助編號  ]NSC82-0408-E110-139
[研究開始日期]1993/02
[研究結束日期]1994/01
[頁 冊 數 ]90頁
[語    文]英文
[建檔日期  ]2000/10
[出版情形  ]完成日期:民國83年1月
[原件提供  ]行政院國家科學委員會科學技術資料中心
NSC82-0408-E110-139
[科資分類號 ]IA0407; IA0202; IA0306
[關 鍵 詞 ]
專家系統; 電腦網路; 網路管理系統; 模糊邏輯; 框架; 物件導向程式規劃;
知識庫; 區域網路

Expert system; Computer network; Network management system; Fuzzy
logic; Frame; Object-oriented programming; Knowledge base; LAN

Computer networks are an essential tool for people in
business, industries, government, and schools. With the rapid rate
of change in network technology and product, and the emergence of
highly sophisticated network users, network design has become an
increasingly complex task. Although the computer society aims at
agreeing to a series of international standards for describing
network architectures, the design of a computer network remains to
be an ill-structured problem which lends itself perfectly to expert
systems solutions. We propose an expert system which is able to
design local area networks meeting the requirements specified by
the user. Rules and guidelines pertaining to local area network
design are formulated and incorporated into the knowledge base. The
system is built on an object-oriented paradigm. The object-oriented
approach and the hierarchical rule structure paradigm are discussed.
We also employ the blackboard technique through which rules can
access dynamic objects conveniently.

[科資中心編號]RB8401-2067
[題    名]
人工智慧與專家系統在台灣森林資源上之應用
Artificial Intelligence and Expert Systems for Forest Resource Management in
Taiwan.
[研 究 者 ]
楊榮啟;林文亮
YANG,YONG-CHI;LIN,WEN-LIANG
[機構名稱  ]台灣大學森林研究所(NTUGFORG)
[經費來源  ]行政院國家科學委員會(NSC)
[補助金額  ]451000
[補助編號  ]NSC82-0409-B002-271
[研究開始日期]1992/08
[研究結束日期]1993/07
[出版日期  ]1993/07
[頁 冊 數 ]53
[語    文]中文
[建檔日期  ]1995/10
[科資分類號 ]GA07;GA0701
[關 鍵 詞 ]
專家系統;疏伐;森林經營

Expert system;Thinning;Forest management

[摘    要]
本年試行建立間伐選木的專家系統。台灣 的森林資源
經營管理特別重視造林事業,現有 人工林面積,除主伐收穫者
外,尚有24萬多公頃, 是振興林業的最佳基礎,因為這些林地多
位在 交通便利地區,伐運費用較低,可供生產木材之 用。又
因其容易達到,可闢做森林遊樂區,供民 眾休閒活動。間伐為
林木從建立到主伐收穫過 程間之重要的撫育兼收收穫作業,
一向受到重 視,理論研究論文及實務發展報告也多,頗值新
進林學家學習。間伐木選定必須考慮經營目的 、生長情況
、鬱閉狀態、輪伐期以及伐運費等 因素,所以有綜合相關研
究及專家的知識與經 驗等而建立專家系統的必要。本專家
系統為林 木經營管理之間伐作業的指導並可以節省調查 費
用。所定之選木方法雖不夠充分尚有待改 進,但將專家系統
應用在間伐選木作業上確是 有突破性的進展。

This year a prototype expert system was developed for the selection of
trees for thinning in the plantations (man-made forests) of the Republic of
China. Since world wide forest resources both in industrialized and developing countries are gradually decreasing, to take a long-term view, it is necessary to maintain and cultivate the national forest resources of one's own country. The foresters in Taiwan, therefore, have a responsibility to develop national forestry industry. The easiest way to achieve this objective is to start from the current physical resources. The one quarter million hectares of plantations in Taiwan are really the best foundation from which to expand our national forestry industry. Most of these plantations are conveniently located and the cost of logging is relatively low for timber production. The plantations can also offer recreation area. Thinning is one of the main silvicultural treatments during the period between the planting and the final harvesting. The purpose of thinning is to improve the quality of the trees that remain in the stand and to acquire monetary benefit from the thinned trees. There are many factors in the selection of trees for thinning such as management purpose, growth condition, crown canopy, rotation, and logging costs. It is essential to develop an expert system by utilizing the studies concerned as well as the knowledge and experience of an exert. The primary objective for the development of this expert system was to provide an advisory system to help forest managers to select trees for thinning in each plantation in Taiwan. Although work is still being done to improve the expert system, the system represents a major step towards improving the thinning procedure in Taiwan.

[科資中心編號]RA8304-2014
[題    名]
專家系統語言編譯器之發展
The Development of an Expert System Language Compiler.
[研 究 者 ]
李錫智
LEE,SHIE-JUE
[機構名稱  ]中山大學電機工程系(NSYEELE)
[經費來源  ]行政院國家科學委員會(NSC)
[補助金額  ]560000
[補助編號  ]NSC82-0408-E110-045
[研究開始日期]1992/08
[研究結束日期]1993/07
[出版日期  ]1993/07
[頁 冊 數 ]104
[語    文]英文
[建檔日期  ]1994/12
[科資分類號 ]IA;IA0305;IA0407
[關 鍵 詞 ]
專家系統;語法;語義學;編譯器;法則網路; 機器學習;人工智


Expert system;Syntax;Semantics;Compiler;Rule network;Machine learning;Artificial intelligence

[摘    要]
專家系統在工業、企業和國防事業上漸趨 重要。在發
展專家系統時,專家系語言是十分 重要的。這個計畫的目的
是實作一個專家系統 語言,該語言的設計是根據國科會計畫
「專家 系統語言的研究」,NSC81-0408-E110-02,我們已發展 一個
新的專家系統語言包含功能基、邏輯基、 法則基等語言的
優點。在我們的語言中,知識 的表示是由Predicate所構成的事
實和法則,而法 則並不一定是Horn Clauses。在事實中可以包含
變 數。對於條件和事實式樣比對是由Predicate logic 中的
Unification達成。執行控制是由Recognize-act cycle和向前鏈結去推
理和解答我們的問題。法 則可以動態的加入或消去而不會
導致知識庫的 不一致性。
我們用LISP實現了以雷梯法則網路為架構 的語言編譯器
。這個編譯器包含了四個主要的 模組:法則編譯模組、事實
解譯模組、衝突解 析模組、行動解譯模組。法則編譯模組
將法則 編譯成法則網路。事實解譯模組進行式樣比對 和部
分比對以收集Activated rules。行動模組執行 定義在被執行法
則RHS的行動。所有的模組用 LISP的函數做出來。法則網路則
是一群LISP中的 全域變數(Golbal variables)。模組間的聯繫是以
函 數呼叫和變數傳輸達成。該語言編譯器經由問 題測試證
明可用。適合用來發展一般應用場合 的專家系統,特別是在
需要具備學習能力的專 家系統上。

Expert systems are increasingly important to industry, business and
defense affairs. Expert system languages are important to the development of expert systems. The purpose of this project is to study expert system
languages and to implement a compiler for the new expert system language that we have designed in the previous project "On Study of New Expert System Languages", NSC81-0408-E110- 02.
We have developed a new expert system language which intends to include the advantages of functional-based, logic-based, and rule- based languages. In our language, knowledge is expressed in the form of facts and rules which consist of predicates, without requiring rules to be Horn clauses. Facts may contain variables. Pattern matching of patterns to facts is done by unification of predicate logic. Control of execution is characterized by the recognize-act cycle of forward-chaining to reason about and answer user questions. Rules can be added and deleted dynamically, without causing any inconsistency in the knowledge base. A compiler for this language has been written in LISP using Rete rule network structure. The compiler consists of four main modules: the rule compiler module, the fact interpreter module, the conflict resolution module, and the action interpreter module. The rule compiler module compiles rules into the rule network. The fact interpreter module performs pattern matching and partial matching for collecting activated rules. The conflict resolution module select a rule to be fired from the activated rules. The action module performs the actions defined in the RHS of the fired rule. All these modules are implemented in LISP functions. The rule network is defined as a set of global variables in LISP. Communication among the modules are done by function calls and variable passing.

[科資中心編號]RA81-2158
[題    名]
具機器學習能力之數位電路板診斷專家系統 (II)
A Digital Circuit Board Diagnosis Expert System with Machine Learning
Capability. (II)
[研 究 者 ]
郭耀煌
KUO,YAU-HWANG
[機構名稱  ]成功大學資訊工程研究所(NCKUIFEG)
[經費來源  ]行政院國家科學委員會(NSC)
[補助金額  ]290000
[補助編號  ]NSC81-0404-E006-120
[研究開始日期]1991/08
[研究結束日期]1992/07
[出版日期  ]1992/08
[頁 冊 數 ]77
[語    文]中文
[建檔日期  ]1994/05
[科資分類號 ]IA;IA0407
[關 鍵 詞 ]
數位電路板故障診斷;專家系統;機器學習; 類神經網路

Digital circuit board fault diagnosis;Expert system;Machine learning; Neural network

[摘    要]
本計畫係「具機器學習能力之數位電路板 診斷專家系
統」第二年計畫,主旨在於開發一 個電路板故障診斷專家系
統,用來找出電路板 上各種故障現象的原因。其最終目標在
於縮短 故障維修所需之時間,使機器故障所造成的損 失能降
到最低點。
第一年計畫的研究成果在於系統架構的設 計、測試輔
助介面的設計、電路之模式方法以 及診斷策略的基本原則
等各方面的研究與探討 。
第二年計畫的研究成果主要是集中在記憶 體副系統測
試,以及ASIC測試樣本之產生等兩大 問題之上。
到目前為止,對於記憶體副系統的測試在 文獻上還未有
很完備的發展,可是記憶體副系 統的測試卻是跟記憶體晶片
的測試具有同等重 要的地位。我們希望能夠在沒有任何外
加設備 及探針(例如邏輯分析儀、針床等)的情況下,藉 分別
測試記憶體副系統中各部分的錯誤來達到 偵測整個記憶體
系統的目的。
本年度的另一個研究重點是ASIC之測試樣 本的產生方法
。我們的作法是先將ASIC的VHDL行 為描述資料轉換成派翠網路
(Petri-net)的表示方 式,再根據此派翠網路對所有可能出現的
錯誤 現象加以處理,利用我們所提出的測試演算法 來產生對
應的測試樣本。

This project intends to develop a digital circuit boards diagnosis
expert system with the machine earning capability, which is to be used in
automatically troubleshooting the various faulty symptoms on circuit boards. The ultimate goal is to shorten the acquired time for troubleshooting, such that the profit is whittled down as less as possible.
In the first year, the main achievements are concentrated in the design
of system architecture, the design of the test-assistant unit, the modeling
methodology, and the basic principles of diagnosis strategy.
In the second year, we focused on other two important problems, which
are hard to handle for those faulty systems with high complexity. They are the testing of the whole memory subsystem and the test pattern generation approach for ASIC.
Although the study of testing for memory subsystems is very poor than
the study of memory chip's testing in the literature, the importance of the
former is not less than the later. Without any supporting of the test
equipments (such as logic analyzer, bed-of-nails), we expected that the whole memory subsystem is fully tested through the testing and the analysis of every sub-units in the memory subsystem. As to the research of the test pattern generation algorithm for ASIC, we firstly translated the corresponding VHDL behavioral description of the concerned ASIC into the equivalent Petri-net representation, then we apply our test pattern generation algorithm directly upon that generated Petri-net.

[科資中心編號]RA81-2131
[題    名]
人工智慧在核能電廠事故診斷上之應用
Artificial Intelligence Applied to System Failure Diagnosis for Nuclear Power
Plant.
[研 究 者 ]
俞偉初
YU,WEI-CHU
[機構名稱  ]行政院原子能委員會核能研究所(INE)
[經費來源  ]中央預算(NB)
[補助金額  ]0
[研究開始日期]1991/07
[研究結束日期]1992/06
[出版日期  ]1991/09
[頁 冊 數 ]74
[語    文]中文
[建檔日期  ]1994/05
[科資分類號 ]IA;IA0407;ID0607
[關 鍵 詞 ]
專家系統;發展環境;可信度因子;前向推論; 後向推論;核能電


Expert system;Development environment;Certainty factor; Forward
chaining;Backward chaining;Nuclear power plant

[摘    要]
本文□述發展中文專家系統發展環境CGW之 目的、其建
立過程及其所具有的特色。我們知 道目前較成熟的專家系
統發展環境都是由國外 引進,欠缺中文的考慮,所以我們發展
「中文專 家系統環境CGW」的目的,就是要提供一個能本 土化
的專家系統發展環境介面,達到(1)知識表 達中文化,(2)推理訊
息顯示中文化,(3)人機介面 中文化,讓使用者能夠運用此發展
環境,迅速建 立專業領域之專家系統,並達成專家系統製作
本土化的需求。
我們的建立過程是考量將國外的一些較成 熟的發展環
境及國內常用的中文系統進行配對 、測試、整合,最後我們
選擇了將GoldWork II及中 文MS-Window結合,此外我們也發展了一
些軟體架 構在此基礎上,這整體我們稱之為CGW。CGW的特 色,
在知識表達方面,我們可用很簡單的語法格 式將專家知識直
接以中文表達出來,且包含可 信度因子的處理;在推理方面,
我們提供前向及 後向推論的方法,自動地將推理、解釋訊息
以 中文直接表達出來,另外我們也提供了與dBASE III的交談方
式。為了方便大家參考,我們將原 始程式及執行過程均附錄
中。

In this paper, we discuss the purpose of the development of "Chinese
expert system development environment CGW", the building process of CGW, and the features of CGW. At present, the popular expert system development tools are developed by foreign companies, so they don't take Chinese characters into account. The purpose of the development of CGW is to offer a native expert system development interface for the following three functions: (1) representing expert knowledge with Chinese characters, (2) displaying the explanations of reasoning process with Chinese texts automatically, (3) showing the man-machine interface with Chinese texts; therefore, users can make use of CGW to develop expert system of some special domain rapidly and to meet the requirements of developing expert system natively. The building process is to evaluate both some foreign mature development tools and some commonly used Chinese systems, and then to select some valuable systems of them to match, test, and integrate each other. Finally, we decide to combine GoldWorks II with MS-Windows, and some software programs based on this structure are developed. We call this integration system "CGW". The features of CGW are as follows: in the respect of knowledge representation, users can express expert knowledge with very simple syntactic formats and Chinese characters. Especially, certainty factor can be included if necessary; in the respect of inference, users can use forward chaining method and backward chaining method to display explanations of inferring process with Chinese texts automatically; besides, CGW has some functions to enable users to interact with dBASE III database. For your reference, we list both the source programs and the process of execution in the appendix.


[科資中心編號]RA8510-2124
[題    名]
模糊專家系統知識擷取環境之研製
Developing a Knowledge Acquisition System for Multiple Experts.
[研 究 者 ]
曾憲雄;黃國禎
TSENG,SHIAN-SHYONG;HWANG,GWO-JEN
[機構名稱  ]交通大學資訊科學研究所(NCTUINSG)
[經費來源  ]行政院國家科學委員會(NSC)
[補助金額  ]408000
[補助編號  ]NSC83-0408-E009-043
[研究開始日期]1994/02
[研究結束日期]1995/07
[出版日期  ]1995/07
[頁 冊 數 ]44頁
[語    文]英文
[建檔日期  ]1997/05
[出版情形  ]研究報告,44頁,民國84年7月
Report, 44p., Jul. 1995
[原件提供  ]行政院國家科學委員會科學技術資料中心
NSC84-2213-E009-066
[科資分類號 ]IA0407
[關 鍵 詞 ]
知識擷取;專家系統;朦朧集合;推論機

Knowledge acquisition;Expert system;Fuzzy set;Inference engine

[摘    要]
在本計畫中,我們研裝出一套以表格為基
底之模糊專家系統發展環境,該環境結合了知
識表格的觀念、模糊理論及個案為基底的推理
方式,它提供了友善的使用者介面、豐富的資
料形態來表達知識、智慧型的知識擷取功能以
及高效率的推理機制。為了使該環境的使用者
更精確的下決策,在推理過程及結論中有三個
因子列入計算,包括適切度、可靠性及確信度,
除此之外,為方便使用者定義適當的關係函數,
系統並提供圖型介面的曲線定義功能來協助模
糊知識的表達。

In this project, a grid-based fuzzy expert system environment,
GBFES, is proposed, which employs the concepts of repertory grid,
fuzzy theory and case-based reasoning. It offers friendly user
interfaces, abundant data types for knowledge representations,
intelligent knowledge acquisition facilities and an efficient
inference engine. To help the users in making precise decisions,
GBFES offers three factors (feasibility, reliability and certainty)
to represent the results of inference. Moreover, GBFES employs a
Bezier curve-based algorithm to help the experts in defining
membership functions.

科資中心編號: RB9102-1001 
題名: 智慧型環境中人物之主動性追蹤與辨識
Active Tracking and Recognition of People under Smart Environments 
研究者: 江政欽 
機構名稱: 東華大學資訊工程系 
經費來源: 行政院國家科學委員會 
補助金額: 新台幣 422 仟元 
補助編號: NSC90-2218-E259-001 
研究日期: 2001/01 - 2001/07 
出版日期: 2001/10 
頁冊數: 5頁 
原件提供: 行政院國家科學委員會科學技術資料中心, NSC90-2218-E259-001 
科資分類號: IA0407;IA0403 
關鍵詞: 人臉辨識 ;人臉追蹤 ;類神經網路 ;智慧型環境 ;電子商務 
Face Recognition ;Face Tracking ;Neural Network ;Smart Environment ;Electronic Commerce 
中文摘要: 近年來,由於計算機計算能力的快速提升以及影像輸入裝置的普及,加上影像處理以及圖型辨識的技術上獲得長足的進步,也因而產生了許多智慧型電腦視覺認知方面的相關應用。例如,最近越來越受到矚目的生物認證、電子商務以及智慧型環境等應用中的人臉辨識即為一例。在一般的人臉辨識系統中,由於環境光線狀態、使用者的臉部角度以及使用者表情等的複雜變化,使得辨識效果經常大打折扣。本計畫的目的即在於針對上述幾個技術困難點進行先期性的研究並嘗試提出改進的可能方案。 
英文摘要: In recent years, the drastic advancement of computer capability and the fast popularization of image acquisition devices had motivated the developments of efficient image processing and pattern recognition techniques. These intelligent technologies bringed about many smart applications of computer vision. One of the many examples is the recognition of human faces, which is getting more and more attractive in applications of biometrics, electronic commerce, and smart environments, etc. Due to the unstable variations of environment lights, orientations of human head and expressions of human faces, the recognition accuracy of most existing systems are usually degraded to an unsatisfactory performance. It is the goal of this project to study these affecting issues and to propose some possible solutions. 

科資中心編號: RB9101-1258 
題名: 半隨機取樣線法與Gabor濾波器應用於人像識別之研究 
Semi-Random Sampling Line Method and Gabor Filter for Face Recognition 
研究者: 陶金旭 
Taur, Jin-Shiuh 
機構名稱: 中興大學電機工程系 
經費來源: 行政院國家科學委員會 
補助金額: 新台幣 312 仟元 
補助編號: NSC89-2218-E005-002 
研究日期: 2000/08 - 2001/07 
出版日期: 2001/07 
頁冊數: 5頁 
原件提供: 行政院國家科學委員會科學技術資料中心, NSC89-2218-E005-002 
科資分類號: IA0406;IA0403;IA0407 
關鍵詞: 人臉辨識 ;Gabor濾波器 ;人臉偵測 ;類神經模糊網路 ;半隨機取樣線法 
Face Recognition ;Gabor Filter ;Face Detection ;Neural Fuzzy Network ;Semirandom Sampling Line Method 
中文摘要: 人像辨識系統是一種方便且有效的生物測定認證系統,它可在出入控制,金融交易,搜尋資料庫等方面有許多實際的應用。近年來有很多人像辨識的相關研究,其中Vel 提出了一個以取樣直線為基礎的人像辨別方法,在此一方法中,以一些隨機取樣的直線,取出劃過人臉影像的亮度值,作為比較的特徵。此一方法,在人臉轉動,以及改變大小之後,都仍有很高的辨識率。但此一方法不能適用於光源有較大變化的情形。在此計畫中,我們以我們已經發展的人像偵測系統為基礎,加入隨機取樣線法,配合使用Gabor wavelets ,使得亮度對人像辨識的影響能夠減小。同時利用經驗法則以減少所需比對的直線數目,以加快辨識的速度,建構出一個可以於複雜環境下進行辨識工作的即時系統。 
英文摘要: Face recognition system is a convenient and effective biometric identification system that can be applied to many practical applications in the areas of access control, automation, and information security. Recently, several face recognition techniques have been proposed. Vel proposed a line-based approach for face recognition, in which a set of random lines is used to exploit the face image. This approach is insensitive to the variations of rotation and the size of the face image. However, it cannot deal with the situation when the lighting condition changes to certain degree. In this proposal, we build an on-line face detection and recognition system, which can work well in the complex environments, based on our previous system. The Gabor functions are adopted for the face recognition system to reduce the adverse effects form the variations of the lighting condition. Moreover, heuristic rules are used to cut the number of the lines required for recognition to increase the speed. 

科資中心編號: RB9012-0062 
題名: 液晶相關濾波器最佳化設計的研究 
Investigation of Optimal Designs for Liquid Crystal Correlation Filters 
研究者: 陳祖龍 
Chen, Chu-Lung 
機構名稱: 元智大學電機工程系 
經費來源: 行政院國家科學委員會 
補助金額: 新台幣 577 仟元 
補助編號: NSC89-2215-E155-008 
研究日期: 2000/08 - 2001/07 
出版日期: 2001/10 
頁冊數: 6頁 
原件提供: 行政院國家科學委員會科學技術資料中心, NSC89-2215-E155-008 
科資分類號: IE0402 
關鍵詞: 液晶相關濾波器 ;印鑑辨識 ;指紋辨識 ;人臉辨識 
Crystal Correlation Filter ;Seal Recognition ;Fingerprint Recognition ;Face Recognition 
中文摘要: 在本計畫中,為了辨識變形的影像,我們將液晶相關濾波器應用在最小平均交叉相關能量的非零階聯合轉換相關器上。我們發現並證明了假若所需的交叉相關峰值被設定為實數,則參考函數也將是實數值的。有個例子證明了辨識的有效性---真正的印章圖像與非常相似印章圖像的區別。利用類似結合了光與電優點的技術,我們可將其應用於指紋辨識系統、人臉辨識系統上,甚至軍事雷達影像偵測上。 
英文摘要: In this project, in order to achieve distortion invariant image recognition, the design of the nonzero order joint transform correlator with minimum average cross-correlation energy has been studied. In this work, we prove that the optimal reference function is real-valued if the desired cross-correlation peak is also set to be real for every image in a training set. The recognition effectiveness is demonstrated by an example, where the genuine seal-imprint can be detected as compared with the invalid, although very similar one. By combining advantages of electronics and optics, we can apply the similar technique to fingerprint recognition system, face pattern recognition, and synthetic aperture radar (SAR) images detection system. 

科資中心編號: RB9102-0995 
題名: 應用影像處理及主動視覺製作頭部位置自動追蹤系統
Automatic Head Pose Tracking System Using Image Processing and Active Vision 
研究者: 毛齊武 
Mao, Chi-Wu 
機構名稱: 成功大學電機工程系 
經費來源: 行政院國家科學委員會 
補助金額: 新台幣 534 仟元 
補助編號: NSC89-2213-E006-153 
研究日期: 2000/08 - 2001/07 
出版日期: 2001/09 
頁冊數: 5頁 
原件提供: 行政院國家科學委員會科學技術資料中心, NSC89-2213-E006-153 
科資分類號: EA0200;EA1501;IA0403 
關鍵詞: 主動式視覺 ;影像處理 ;人臉辨識 ;橢圓形頭部追蹤 
Active Vision ;Image Processing ;Face Recognition ;Elluptical Head Tracker 
中文摘要: 因為人臉辨識、眼控系統、數位影像的編碼以及視訊會議等的應用在近幾年來需求日殷,所以人臉追蹤及方向的判別越來越重要。我們的研究是建立一個人臉追蹤及方向判別的系統,目的是要以影像處理的技術與幾何關係在連續的影像中自動追蹤人臉及判斷人臉的方向,並使其具有偵測錯誤及即時處理的能力。在人臉追蹤方面,會用一簡單而快速的橢圓形頭部追蹤演算法,增加一些影像的前置後,改良成我們所需要的人臉追蹤系統。人臉方向判斷方面,會利用我們建立的眼睛追蹤演算法來追蹤雙眼,並把雙眼假設為圓球表面的兩點,以眼部影像的中心點來作為球體的中心點,最後從這三點構成的關係,求出人臉的方向。 
英文摘要: Face tracking and pose estimation is getting more and more important recently because it can be applied in a variety of applications, e.g. the face recognition, eye gaze control system, video coding and teleconference, etc. We have recently established a face tracking and pose estimation system in our research. The purpose of this system is to have the computer detect the human face and estimate the pose automatically through the digital process and some relationship of geometry. The system has the ability of fault detection and it can process in real time. We use a simple algorithm called elliptical head tracker to track a person’s head, adding some preprocess to improve the performance for our face tracking system. We also developed an algorithm for tracking the eyes robustly. In order to estimate the direction of the face, we assume the eyes are on the surface, we assume the eyes on the surface of a sphere and the image of the face center is the center of that sphere. So we can compute the direction of the surface that is composed of these points. 

科資中心編號: RB9010-0070 
題名: 一個階層式搜尋架構的人像識別系統
A Face Recognition System via Hierarchical LDA-Based Search Approach 
研究者: 廖弘源 
Liao, Hong-Yuan 
機構名稱: 中央研究院資訊科學研究所 
經費來源: 行政院國家科學委員會 
補助金額: 新台幣 467 仟元 
補助編號: NSC88-2213-E001-017 
研究日期: 1998/08 - 1999/07 
出版日期: 1999/07 
頁冊數: 5頁 
原件提供: 行政院國家科學委員會科學技術資料中心, NSC88-2213-E001-017 
科資分類號: IA0403;IA0406 
關鍵詞: 人臉辨識 ;線性鑑別分析 ;階層式分類 ;回溯搜尋 ;特徵向量 
Face Recognition ;Linear Discriminant Analysis (Lda) ;Hierarchical Classification ;Backtracking Search ;Feature Vector 
中文摘要: 在本計畫中我們提出一個以LDA為基礎的階層式人臉辨識系統。我們將人臉影像按不同解析度轉換到波元範圍。因為不同解析度將產生不同長度之特徵向量,我們利用較粗糙的特徵向量去執行快速篩選。因為粗糙的向量可以快速剔除大部份不合適的人臉影像。接著,我們使用較精確細密的特徵向量執行細部比對。實驗結果証明我們的方法的確可行,而且比文獻上記載類似的方法有效率且正確度高?C 
英文摘要: In this project, we have proposed a face recognition system that bases its search on a hierarchical LDA-based search tree. First, the input image was converted to several feature vectors of multi- resolution format by wavelet transforms. The coarse feature vectors were used to efficiently delete most of impossible candidates. Then, the fine feature vectors were used to correctly identify the candidate set in the database. In addition, an evaluation procedure was performed to evaluate the similarity score between the query image and the candidate set. If the similarity score is low, a backtracking process was triggered to correct the mis-classified items and then improve the recognition rate. Some experimental results were conducted to show the efficiency and effectiveness of our approaches. 

科資中心編號: RB8910-0343 
題名: 類神經網路靜態人像辨識系統之研究(III) 
The Study of a Neural Network Face Recognition System for Static Images (III) 
研究者: 傅心家 
Fu, Hsin-Chia 
機構名稱: 交通大學資訊工程系(NCTECSIE) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 426 仟元 
補助編號: NSC88-2213-E009-053 
研究日期: 1998/08 - 1999/07 
原件提供: 行政院國家科學委員會科學技術資料中心 NSC88-2213-E009-053 
科資分類號: IA0406;IA0407 
關鍵詞: 人臉辨識 ;類神經網路 ;靜態影像 ;特徵選擇 ;圖形辨識 
Face Recognition ;Neural Network ;Static Image ;Feature Selection ;Pattern Recognition 
中文摘要: 選擇具代表性的特徵在人像辨識上是一件重要的工作,傳統的人像辨識方法大多使用相同的特徵集合來區別不同的人,而這樣的限制相信是可以被打破的。我們採用了每個人像使用不同的特徵集合的方式來提高辨識率。在本計劃中,藉由一個特徵選取的演算法刪除了每個人像中最不重要或容易產生混淆的特徵,僅僅保留能夠有效區別不同人像的特徵,以提供機率抉策神經網路(PDBNN)做人像辨識之用。我們將所發展出的方法分別應用到兩個不同的人像資料庫上,都得到不錯的實驗結果。在本計劃所建立的151人的人像資料庫中,辨識率可由原來的86.26% (每個人像使用2500個特徵)提高到94.87% (每個人像使用1000個特徵),而在ORL人像資料庫的實驗上,可由原來的92.5% (每個人像使用10304個特徵)提高到98.5% (每個人像使用8000個特徵)。目前如上述辨識功能之雛型展示已完成,請進入:http:// www.csie.nctu.edu.tw/.sim.nnlab瀏覽。 
英文摘要: Selecting proper and important features is an essential task for efficient face recognition. Most of recognition or classification algorithms proposed to extract and to use features in uniform manner from each object. We believe the constraint on uniform number of features could be relaxed to non-uniform and yet to achieve better recognition performance. In this project, we present a PDBNN based feature reduction algorithm that deletes some feature vectors which contribute the least among of the whole feature set. The deletion is performed on individual facial basis. By applying the proposed algorithm, we performed some face recognition experiments on a in-house 151 people facial database and the ORL database. The experimental results show the recognition accuracy improved from the original 86.26% (2500 features) to 94.87% by using only 1000 features on the in-house database, and the recognition accuracy improved from 92.5% (10304 features) to 98.5% (8000 features) on the ORL database. 

科資中心編號: RB8910-0342 
題名: 類神經網路於生物測定認證技術及應用之研究(III) 
The Study of Biometric Identification Technology and Applications by Neural Networks (III) 
研究者: 傅心家 ; 劉長遠 ; 陶金旭 ; 劉啟民 ; 林道通 
Fu, Hsin-Chia ; Liou, Cheng-Yuan ; Taur, Jin-Shiuh ; Liu, Chi-Min ; Lin, Daw-Tung 
機構名稱: 交通大學資訊工程系(NCTECSIE) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 339 仟元 
補助編號: NSC88-2213-E009-052 
研究日期: 1998/08 - 1999/07 
原件提供: 行政院國家科學委員會科學技術資料中心 NSC88-2213-E009-052 
科資分類號: IA0407;IA0406 
關鍵詞: 人臉辨識 ;類神經網路 ;生物測定認證 ;人像偵測 ;形變 
Face Recognition ;Neural Network ;Biometric Identification ;Face Detection ;Morphing 
中文摘要: 應用電腦視覺於生物測定認證系統近來受到越來越多的注意,而生物測定認證在自動化系統之資訊安全及防護上能有許多的實際應用。本計劃中使用多種不同的類神經網路技術,發展出一高效能的生物測定認證系統。基本上,此系統利用視訊信號偵測人像大概位置後,再使用人像膚色的獨特性正確定位出人臉,然後交由靜態人像辨識系統執行辨識以確定是否為合法的使用者,如系統無法辨識時則可將影像交由臉部表情處理系統對臉部表情進行正規化處理,以做進一步的辨識。目前如上述辨識功能之雛型展示已完成,請進入:http://www.csie.nctu.edu.tw /.sim.nnlab 瀏覽。 
英文摘要: The feasibility of using computer vision to identify biometrics has received increasing attention. Biometric identification techniques can be applied to many practical applications in the area of automation and information security. This project has combining various techniques of Neural Network to develop a fully functional biometrics identification system. This system consists of the following subprojects: (1) Static face recognition, (2) Detect human faces in a sequence of images, (3) Facial expression and morphing. The system detects the face location approximately in a sequence of images and then locates the face by the skin color information. A static face recognition subsystem is employed to recognize the located face and judge the person is a legal user or not. If the system fail to recognize the face, a face morphing subsystem then used to normalize the face expressions for recognizing again. 

科資中心編號: RB8910-0340 
題名: 類神經網路在視訊信號中人像偵測與辨識之應用(III) 
Face Detection and Recognition for Video Signal Using Neural Networks(III) 
研究者: 陶金旭 
Taur, Jin-Shiuh 
機構名稱: 中興大學電機工程系(NCHEELE) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 439 仟元 
補助編號: NSC88-2213-E005-012 
研究日期: 1998/08 - 1999/07 
原件提供: 行政院國家科學委員會科學技術資料中心 NSC88-2213-E005-012 
科資分類號: IA0406;IA0407 
關鍵詞: 人臉辨識 ;類神經網路 ;視訊信號 ;人像偵測 ;生物測定認證 ;模糊網路 
Face Recognition ;Neural Network ;Video Signal ;Face Detection ;Biometric Identification ;Fuzzy Network 
中文摘要: 生物測定認證在自動化系統之資訊安全及防護上能有許多實際的應用。此外,近年來由於電腦技術快速進步,更使得生物測定認證可成為一種安全好用的方法。在許多生物測定認證系統中,人像辨識是十分方便並且較不需要使用者作特殊配合的一種方式。本計畫在三年內完成一線上人像偵測與辨識系統,其中包含下列工作:(1) 建立視訊資料庫,(2) 設計人像偵測與辨識的特徵向量,(3) 研究演算法,(4)撰寫影像控制介面及人機介面。目前人像偵測與辨識演算法在Pentium II 400之個人電腦上執行約需1.67秒鐘。在25人的小型資料庫時,準確率可達97.2%以上。 
英文摘要: Biometric identification techniques can be applied to many practical applications in the area of automation and information security. Moreover, the computer technology has advanced rapidly recently. It will make the biometric identification a safe and useful approach. Among the biometric identification methods, the face recognition is a very convenient approach and it does not require the user to take certain action. In the 3-year project, we have developed a on-line face detection and recognition system, which includes the following tasks: (1) building image video databases, (2) selecting the feature vector for face detection and recognition, (3) developing algorithms, and (4) designing the human-machine interface. So far, the whole face detection and recognition algorithm takes about 1.67 second to finish the tasks on a personal computer with Pentium II 400. The error rate is below 2.8% when tested on a small database with 25 persons. 

科資中心編號: RB8910-0327 
題名: 雙眼彩色臉孔辨識之研究 
A Color Face Recognition System Using Binocular Images 
研究者: 陳玲慧 
Chen, Ling-Hwei 
機構名稱: 交通大學資訊科學研究所(NCTUINSG) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 559 仟元 
補助編號: NSC88-2213-E009-063 
研究日期: 1998/08 - 1999/07 
原件提供: 行政院國家科學委員會科學技術資料中心 NSC88-2213-E009-063 
科資分類號: IA0406;IA0402 
關鍵詞: 人臉辨識 ;雙眼影像 ;立體視覺模型 ;特徵點 ;彩色 ;三度空間 
Face Recognition ;Binocular Image ;Stereo Vision Model ;Feature Point ;Color ;Three Dimension 
中文摘要: 在研究計畫中我們提出一種利用立體視覺模型所做的彩色臉孔辨識方法。首先,將兩個攝影機並排以達成如人眼般的視覺模型,並對攝影機前的人取兩張影像。藉著由各個影像中找到特徵點並將兩張影像的相同特徵點做對應,可以算出特徵點的三度空間位置。辨識所需的主要特徵值都是從這些位置所導出的,另外還加上由區域邊緣點所導出的不變量。因為使用三度空間位置,可以避免做大小的調整,也可以避免因為投影所造成的誤差。本系統的訓練過程中含有特徵選取功能,可以選擇分離度較大的特徵用於辨識。本系統的辨識方法是估計不明人物是資料庫中某人的機率。除此之外,為了容易研究與使用,我們為本系統發展了一個方便的PC上的使用者介面。 
英文摘要: In this project, a color face recognition system using stereo vision model is proposed. Two cameras are calibrated to be parallel and an image pair is captured from a person. 3D positions of feature points on face are first found, the feature vector including distances and ratio values derived from feature points and invariant moments calculated from edge points are then extracted. In the training process, a feature selection algorithm is provided to choose features with higher disjunction degree. The recognition process estimates the a posteriori probability of the unknown belonging to each person. A user friendly interface is developed on PC for convenient research and use. 

科資中心編號: RB8803-0152 
題名: 類神經網路於生物測定認證技術及應用之研究(II)(總計畫) 
The Study of Biometric Identification Technology and Applications by Neural Networks (II) 
研究者: 傅心家 ; 陶金旭 
Fu, Hsin-Chia ; Taur, Jin-Shiuh 
機構名稱: 交通大學資訊工程系(NCTECSIE) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 127 仟元 
補助編號: NSC87-2213-E009-040 
研究日期: 1997/08 - 1998/07 
頁冊數: 4頁 
原件提供: 行政院國家科學委員會科學技術資料中心 NSC87-2213-E009-040 
科資分類號: IA0407;IA0404;IA0406 
關鍵詞: 生物認證 ;類神經網路 ;人臉辨識 ;唇語判讀 ;語音辨識 ;形變 
Biometric Identification ;Neural Network ;Face Recognition ;Lip Reading ;Speech Recognition ;Morphing 
中文摘要: 運用生物認證(biometrics identification)技術來做身分識別目前已成為一件具有廣大市場需求以及挑戰性的工作。生物認證包括了指紋識別、語音識別、簽名識別、眼彩虹識別,以及人像識別等等。本計畫預計將完成一結合靜態人像辨識、視訊人像辨識、唇語判讀、語音及語者辨識,與人像表情處理辨識等的整合型生物測定認證系統。本計畫為三年計畫中之第二年,在本年度的計畫工作中,我們已完成不同子系統間的輸出入介面設計,同時五個子計畫已實作完成可獨立運作的五個生物測定認證子系統,然後預計在第三年的計畫中,我們將整合五個子系統,實作完成一整合型的生物測定認證系統。由於此系統結合了人像、唇語及語音的交互認證,因此此一整合系統比任何單一功能的生物認證系統具有更佳的防偽功能,除此之外尚具有非強迫性(non-intrusive)的人性化優點,亦即使用者僅需站在系統前,念出屬於自己的 ID,即能達到身分認證的目的。相對於指紋辨識系統必須以手指觸摸指紋感知器,或眼彩虹辨識系統必須以特殊光線照射使用者瞳孔以讀取資訊,此一整合影像及語音的生物測定認證系統具有加諸受測使用者最少負擔的人性化優點。 
英文摘要: Recently biometrics identification becomes more and more popular in both academic and commercial areas. Biometrics identification includes such as finger prints, speech recognition, signature dynamics , iris recognition, and face recognition. This project intends to construct a biometrics identification system that involves static face recognition, video face recognition, lips reading, speech recognition, and facial expression morphing systems. This is the second year of three-year project. During this year we have design the input/output interfaces among the subsystems such as data transference protocol, data exchange format. Additionally, five standalone subsystems have been implemented in the end of this year. Then, these subsystems will be fused to form a integrated biometrics identification system in the third year of project. Due to the ability of mutual identification by lips reading, face and speech recognition subsystems, the integrated biometrics identification system is better in anti-mask than other systems and has the benefit of being a passive, non-intrusive system for verifying personal identity. The user to be verified needs only to stand in the front of system and says one's personal ID without any other actions. Contrast to the finger prints system that needs the user to touch the sensor, or the iris recognition system that needs the user pupil of the eye to be shined by special light. The biometrics identification integrated system that combine face images and voices are friendly as well as lesser constraint to the user. 

科資中心編號: RB8803-0151 
題名: 類神經網路於生物測定認證技術及應用之研究(II)---子計畫I:類神經網路靜態人像辨識系統之研究(II) 
The Study of a Neural Network Face Recognition System for Static Images (II) 
研究者: 傅心家 
Fu, Hsin-Chia 
機構名稱: 交通大學資訊工程系(NCTECSIE) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 598 仟元 
補助編號: NSC87-2213-E009-041 
研究日期: 1997/08 - 1998/07 
頁冊數: 4頁 
原件提供: 行政院國家科學委員會科學技術資料中心 NSC87-2213-E009-041 
科資分類號: IA0406;IA0407 
關鍵詞: 人臉辨識 ;類神經網路 ;混成系統 ;特徵臉 ;大眾臉 
Face Recognition ;Neural Network ;Hybrid System ;Eigenface ;Popular Face 
中文摘要: 在本年的計畫中我們完成了兩類型的靜態人像辨識系統:混合型人像辨識系統與多專家系統。混合型人像辨識系統,應用特徵臉人像辨識系統[1]挑選出候選的相似臉,再由其後所串接的機率決策神經網路[2]決定最後的辨識結果。而多專家系統訓練一些辨識器去辨識不同角度的人像,再由一個Gating network[3]去整合這些辨識器的辨識結果。我們的實驗顯示,這兩種辨識系統都能比特徵臉辨識系統達成更高的辨識率。 
英文摘要: In the past year we have developed two face recognition models for recognizing static face images: the hybrid face recognition system and the multi-expert face recognition system. The hybrid face recognition system adopts an eigenface [1] subsystem selecting similar faces to target first, then pass these similar faces to a probabilistic decision based neural network [2], to identify the target . The multi-expert system recognizes various orientation faces by the dedicated expert recognizers then combines the recognition results by a gating network [3]. The experiment results show that our face recognition systems achieve better recognition performance than the eigenface system. 

科資中心編號: RB8801-0032 
題名: 利用形態學前處理做快速人臉偵測
Fast Face Detection via Morphology-Based Preprocessing 
研究者: 廖弘源 
Liao, Hong-Yuan 
機構名稱: 中央研究院資訊科學研究所(ASII) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 356 仟元 
補助編號: NSC87-2213-E001-025 
研究日期: 1997/08 - 1998/07 
頁冊數: 25頁 
原件提供: 行政院國家科學委員會科學技術資料中心 NSC87-2213-E001-025 
科資分類號: IA0406 
關鍵詞: 人臉偵測 ;形態運算 ;類神經網路 ;人臉識別 
Face Detection ;Morphological Operation ;Neural Network ;Face Recognition 
英文摘要: An efficient face detection algorithm which can detect multiple faces in a cluttered environment is proposed. The proposed system consists of three main steps. In the first step, a morphology-based technique is devised to perform eye-analogue segmentation. Morphological operations are applied to locate eye-analogue pixels in the original image. Then, a labeling process is executed to generate the eye-analogue segments. In the second step, the previously located eye-analogue segments are used as guides to search for potential face regions. The last step of the proposed system is to perform face verification. In this step, every face candidate obtained from the previous step is normalized to a standard size. Then, each of these normalized potential face images is fed into a trained backpropagation neural network for identification. After all the true faces are identified, their corresponding poses are located based on guidance obtained by optimizing a cost function. The proposed face detection technique can locate multiple faces oriented in any directions. Furthermore, the morphology-based eye-analogue segmentation process is able to reduce the background part of a cluttered image by up to 95%. This process significantly speeds up the subsequent face detection procedure because only 5-10% of the regions of the original image remain for further processing. Experiments demonstrate that an approximately 94% success rate is reached, and that the relative false detection rate is very low. 

科資中心編號: RB8707-0478 
題名: 類神經網路於生物測定認證技術及應用之研究(I):子計畫(一)---類神經網路靜態人像辨識系統之研究 
The Study of a Neural Network Face Recognition System for Static Images (I) 
研究者: 傅心家 
Fu, Hsin-Chia 
機構名稱: 交通大學資訊工程研究所(NCTUCSIG) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 495 仟元 
補助編號: NSC86-2745-E009-009 
研究日期: 1996/08 - 1997/07 
頁冊數: 45頁 
原件提供: 行政院國家科學委員會科學技術資料中心 NSC86-2745-E009-009 
科資分類號: IA0407;IA0406 
關鍵詞: 人臉辨識 ;類神經網路 ;特徵臉 ;人像資料庫 ;主要元素分析 
Face Recognition ;Neural Network ;Eigenface ;Face Database ;Pca 
中文摘要: 運用生物認證技術來做身分識別目前已成為一件具有廣大市場需求以及挑戰性的工作。本子計畫預計將完成一靜態人像辨識系統。在過去的一年中,我們已建構一包含1079人的大型靜態人像資料庫系統,以作為發展人像辨識系統的實驗測試之用。前述建構之人像資料庫已完成,原則上歡迎學者專家利用做學術研究,唯各影像涉及肖像權,目前正會同國科會智慧財產權專家,研擬使用規則,等規則制定後自將公告,並歡迎大家利用。我們擬採用混合型人像辨識系統,應用特徵臉人像辨識系統挑選出候選的相似臉,再由其後所串接的機率決策神經網路決定最後的辨識結果。這種混合系統的目的在於融合不同人像辨識方法的優點,截長補短,達到最佳的效果。在加快辨識時間方面,我們提出結合機率決策神經網路與主要特徵分類法的架構,利用循序尋找主要高斯特徵,並且移除已被主要特徵正確分類的訓練資料,由於主要特徵能捕捉大部分訓練/測試資料的分布,因此除可以加快訓練速度外,也可降低辨識所需的平均時間。 
英文摘要: Recently biometric identification becomes more and more popular in both academic and commercial areas. This project intends to construct a static face recognition system. For the past year we have created a static facial database of 1079 persons for the use of developing and testing static face recognition system. In general, we would like to open this face image database for academic use. However, every face image or picture is associated with individual portrait property, we are working with intellectual property expert to come out some rules of using this database. As long as these rules are ready we will announce to the public and welcome everyone interested to share this database. We intend to solve the face recognition problem by adopting a hybrid structure which an eigenface subsystem selecting similar faces to target first, then pass these similar faces to a probabilistic decision based neural network, to identify the target. This type of hybrid system try to combine the advantageous of two different systems and to even up disadvantageous for better recognition rate. We also proposed a neural network which combine the advantageous of PDBNN and the principal feature classification for reduce the average recognition time. The network is designed by sequentially finding principal Gaussian features and removing training data which has already been correctly classified. Since the principal Gaussian features model the main distribution of training/ testing data, the corresponding training/testing time can be great saved. 

科資中心編號: RB8705-0181 
題名: 類神經網路/隱藏式馬可夫模型混合系統在臉部辨識的研究 
The Study of a Hybrid Neural Network/Hidden Markov Model System on Face Recognition 
研究者: 陳嘉琳 
Chen, Chia-Lin 
機構名稱: 中華工學院資訊工程系(CHPICOS) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 389 仟元 
補助編號: NSC86-2213-E216-019 
研究日期: 1996/08 - 1997/07 
頁冊數: 76頁 
原件提供: 行政院國家科學委員會科學技術資料中心 NSC86-2213-E216-019 
科資分類號: IA0403 
關鍵詞: 影像辨識 ;人臉辨識 ;自組織特徵圖 ;類神經網路 ;隱藏式馬可夫模型 
Image Recognition ;Face Recognition ;Self-Organizing Feature Map ;Neural Network ;Hidden Markov Model 
中文摘要: 「臉部影像分析」的問題,是跨領域的一個熱門研究題材。臉部影像若能透過電腦處理分析,則可以取代原本須以人類肉眼方能完成的反覆而且繁瑣的工作,也有助於了解人類心理的情緒反應。而在眾多的臉部影像分析問題中,「個人的辨識」是臉部影像分析中的最基本的工作,而對此問題的了解,有助於其他問題的研究。本報告中的研究,是使用統計模型建立一個臉部模型,並探討此模型所適用的臉部分析問題。臉部影像的分析問題,通常是基於臉部的特徵來完成。一般而言,巨觀性的特徵較能容忍臉部影像巧妙變化的差異,而微觀性的特徵較能得到準確的局部臉部訊息,但其特徵不容易定義,其特徵訊息也會因特徵的不同而有差異。臉部辨識的問題,是微觀性亦是巨觀性,必須同時兼顧此雙面性,方能解決臉部此具巧妙變化性的問題。在本研究中,我們便是基於此論點,將臉部影像做微觀性的分析,再做巨觀性的整合,以此做為臉部影像的模型。本計畫今年完成的目標如下:(1)臉部分割問題:此為分解臉部影像,以利於進行微觀性的分析。(2)影像編碼問題:基於類神經網路,以編碼的方式,進行特徵的定義及訊息的提取。(3)訊息整合問題:此為使用雙層隱藏式馬可夫模型,整合編碼後之分割後影像的訊息,分為縱向與橫向的整合。(4)分析此模型適用的臉部分析問題。 
英文摘要: The problem of face analysis is a popular inter-discipline research topic. In case that the computer owns the capability of automatic face analysis, the trivial and complicated work of verification then can be done easily by computers. Besides, the emotional mode of human being can also be easily interpreted by analyzing the faces. Among the various face-related problems, face recognition is the most essential and the solution to this problem is beneficial to other face related problems. In this report, we study a model for the general face problems, instead of a scheme for a particular purpose. Based on the artificial neural network (ANN) and hidden Markov model (HMM), we build a statistical model for the face image. Usually, the analysis of face image is based on the facial features. Generally speaking, the global features can tolerate the local changes of features such as expression. On the other hand, the local features can capture more accurate facial information though, it is a very tuff to define appropriately. Obviously, a proper face model should be based on the local facial information as well as the global one. In this project, we propose a hybrid model to locally analyze the facial images and globally integrate the local information. In this project, we have accomplished: (1) Decomposition of facial image for the purpose of local analysis. (2) Based on an ANN to encode the partitioned facial images. In this way, we define and explore the facial features locally in a statistical way without"hard" definition of features. (3) We propose a double HMM to integrate the encoded information of partitioned images computed. The integration is performed in two layers, first horizontally then vertically. (4) The applicability of this model to the problems of face analysis. 

科資中心編號: RA9010-0031 
題名: 灰階臉孔辨識之研究(I) 
A Study on Gray Scale Human Face Recognition(I) 
研究者: 陳玲慧 
Chen, Ling-Hwei 
機構名稱: 交通大學資訊科學系 
經費來源: 行政院國家科學委員會 
補助金額: 新台幣 553 仟元 
補助編號: NSC85-2213-E009-124 
研究日期: 1995/08 - 1996/07 
出版日期: 1996/07 
頁冊數: 45頁 
原件提供: 行政院國家科學委員會科學技術資料中心, NSC85-2213-E009-124 
科資分類號: IA0406 
關鍵詞: 人臉辨識 ;人臉定位 ;特徵臉 ;紋理 ;馬賽克影像 ;影像分割 
Face Recognition ;Face Identification ;Eigenface ;Texture ;Mosaic Image ;Image Segmentation 
中文摘要: 本計劃提出一套自動的人臉定位及切割方法,它主要是根據人臉部五官的對稱性,先找出人臉的對稱中線,再以這條對稱中線為基礎,並根據頭髮與眼睛的相關位置以及它們在影像中的灰階值特性,找出眼睛的位置,然後依據人雙眼的兩個眼珠之間距離固定的假設,定位並進而切割出臉部的區域。由於切割出來的影像只包含臉部,因此可以作為人臉辨識系統的輸入影像,根據實驗結果,所提出的定位切割方法在不同的光線、背景及距離等因素下都有很高的正確率,證明此方法的可行性。 
英文摘要: In this project, an automatic face location algorithm will be proposed. The algorithm is based on the symmetry of a face and the reltive position between hair and eyes. The characteristics of the gray levels of hair and eyes are used to clip the face region out of an image. The proposed approach finds the center line of symmetry of the face in an image. Then the hair region is estimated and the rough size of the face is known. The gray level variations and the symmetry property of a face are used to find the position of eyes. After the eye positions are located, the face region is located and clipped out of the image. 

科資中心編號: RB8707-0452 
題名: 灰階臉孔辨識之研究(I) 
A Study on Gray Scale Human Face Recognition (I) 
研究者: 陳玲慧 
Chen, Ling-Hwei 
機構名稱: 交通大學資訊科學研究所(NCTUINSG) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 553 仟元 
補助編號: NSC85-2213-E009-124 
研究日期: 1995/08 - 1996/07 
頁冊數: 43頁 
原件提供: 行政院國家科學委員會科學技術資料中心 NSC85-2213-E009-124 
科資分類號: IA0406 
關鍵詞: 人臉辨識 ;影像處理 ;人臉定位 ;紋理 ;馬賽克 ;特徵臉 
Face Recognition ;Image Processing ;Face Identification ;Texture ;Mosaic Image ;Eigenface 
中文摘要: 本計畫提出一套自動的人臉定位及切割方法,它主要是根據人臉部五官的對稱性,先找出人臉的對稱中線,再以這條對稱中線為基礎,並根據頭髮與眼睛的相關位置以及它們在影像中的灰階值特性,找出眼睛的位置,然後依據人雙眼的兩個眼珠之間距離固定的假設,定位並進而切割出臉部的區域。由於切割出來的影像只包含臉部,因此可以作為人臉辨識系統的輸入影像,根據實驗結果,所提出的定位切割方法在不同的光線、背景及距離等因素下都有很高的正確率,證明此方法的可行性。 
英文摘要: In this project, an automatic face location algorithm will be proposed. The algorithm is based on the symmetry of a face and the relative position between hair and eyes. The characteristics of the gray levels of hair and eyes are used to clip the face region out of an image. The proposed approach finds the center line of symmetry of the face in an image. Then the hair region is estimated and the rough size of the face is known. The gray level variations and the symmetry property of a face are used to find the position of eyes. After the eye positions are located, the face region is located and clipped out of the image. 

科資中心編號: RB9102-0358 
題名: 具容錯能力的會議金匙系統之研究與製作 
Study on Conference Key Agreement Systems with Fault Tolerance 
研究者: 曾文貴 
Tzeng, Wen-Guey 
機構名稱: 交通大學資訊科學系 
經費來源: 行政院國家科學委員會 
補助金額: 新台幣 373 仟元 
補助編號: NSC89-2213-E009-180 
研究日期: 2000/08 - 2001/07 
出版日期: 2001/07 
頁冊數: 6頁 
原件提供: 行政院國家科學委員會科學技術資料中心, NSC89-2213-E009-180 
科資分類號: IA0101;IA0202 
關鍵詞: 會議金匙 ;容錯 ;安全多方計算 ;金匙協議 ;網際網路 ;密碼協定 
Conference Key ;Fault Tolerance ;Secure Multiparty Computation ;Key Agreement ;Internet ;Cryptographic Protocol 
中文摘要: 當一群人想要在開放式的網路上召開一個會議時,為了安全,他們應該先建立起一把共同的會議金匙,再用此會議金匙加密所有的通訊內容,以保障安全。然而在建立會議金匙的過程中,可能有部份的參與者惡意地欺騙其他的參與者,使得誠實的參與者不能得到共同的會議金鑰值。本計畫研究一個在沒有可信賴中心存在的分散式環境下,由所有參與者共同建立一會議金鑰的協定。且在惡意的參與者少於一半(或者任意多)的情況下,所有誠實的參與者可以建立起唯一的會議金鑰值,同時以零知識證明的方式證明此協定對於在一旁竊聽的旁觀者而言,將不會洩露出任何的資訊。 
英文摘要: When a group of members want to hold a conference over Internet, they should establish a common conference key so that the attackers cannot get the information of their communication. We consider not only the eavesdroppers, but also malicious members who try to spoil the conference sot that honest members cannot have a common conference key. We proposed a distributed conference key agreement system. In our system, honest members can compute a common key no matter how many members are malicious. We also showed that an eavesdropper can not compute any information about the key. 

科資中心編號: RB9102-0316 
題名: 在分散式物件導向系統中容錯垃圾收集問題之研究 
The Study of Distributed Object-Oriented Fault-Tolerant Garbage Collection Problem 
研究者: 蕭顯勝 
Siu, Hin-Sing 
機構名稱: 台灣師範大學工業科技教育系 
經費來源: 行政院國家科學委員會 
補助金額: 新台幣 296 仟元 
補助編號: NSC89-2213-E003-010 
研究日期: 2000/08 - 2001/07 
出版日期: 2001/07 
頁冊數: 4頁 
原件提供: 行政院國家科學委員會科學技術資料中心, NSC89-2213-E003-010 
科資分類號: IA0301;IA0201 
關鍵詞: 分散式物件導向系統 ;容錯 ;垃圾收集 ;分散式協議 ;混合錯誤模式 
Distributed Object Oriented System ;Fault Tolerance ;Garbage Collection ;Distributed Agreement ;Hybrid Fault Model 
中文摘要: 在實際的情況中,大部份網路的連接方式都不是全連接的(Fully connected)。而且,處理機和通訊線路都會發生不同的錯誤模式,稱為混合錯誤模式(Hybrid fault model)。在本計畫中,我們將對垃圾收集問題在混合錯誤模式下作研究。我們在本計畫中提出了一個垃圾收集問題之演算法FTGS ,此演算法針對分散式系統物件能分佈在每一個處理器的情況下作垃圾收集。當每一個處理器執行完 FTGS 後即可將分散在系統中之垃圾(可重新使用之記憶體)重新使用。FTGS 能容忍最大數目的錯誤單元(處理機和通訊線路),使得所有功能良好處理機在一般性網路中都能解決垃圾收集之問題。 
英文摘要: In distributed systems, actually, the processors can be subjected to different types of fault simultaneously, also referred to as hybrid fault model. The problem of distributed garbage collection is revised under hybrid fault model. An algorithm is proposed for the garbage collection in distributed systems with object sharing across the memory space of the processors. The proposed algorithm is executed with fully asynchronous manner at each processor. It requires no global synchronization or the status of the processors in the system. The inaccessible objects can be reclaimed as free space after the algorithm is executed. Finally, the correctness and complexity of the algorithm are discussed. 


科資中心編號: RN9102-0134 
題名: 嵌入式環狀網路最佳多容錯性質之研究 
The Optimal K- Fault-Tolerance for Embedding Ring Networks 
研究者: 洪春男 
機構名稱: 大葉大學資訊工程系 
經費來源: 行政院國家科學委員會 
補助編號: NSC89-2218-E212-002 
研究日期: 2000/08 - 2001/07 
出版日期: 2001/10 
頁冊數: 5頁 
原件提供: 行政院國家科學委員會科學技術資料中心, NSC89-2218-E212-002 
科資分類號: IA0202;IA0101 
關鍵詞: 容錯 ;互連網路 ;環狀網路 ;嵌入 ;Pancake網路 ;漢米爾頓迴圈 
Fault Tolerance ;Interconnection Network ;Ring Network ;Embedding ;Pancake Network ;Hamiltonian Circuit 
中文摘要: 一個連結網路連結了平行電腦上的處理器,而環狀網路是最基本的一種連結網路,環狀網路常被用來當作區域網路的連接結構,例如號誌環網路(Token ring)。圖形嵌入問題(Graph embedding problem)是衡量連結網路最中心的主題之一,給定一個圖形G 與另一個主要圖形H 這個問題是找到一種將圖形G中的任何一點對映到H 中的一個點,而G 中的任何一邊對映到H 中的某條路徑(在我們的研究中主要對映到H 中的一個邊)容錯( Fault tolerance )問題也是一個很重要的主題,特別是當連結網路所連結之處理器個數增加時,在這個計劃中,我們在某些重要連結網路上作邊與點容錯之嵌入式環狀網路之研究 。一個圖形可以用來表示連結網路,用圖形上的點來表示處理器,並用邊來表示處理器之間的連結。用圖形G = (V, E) 來表示一個連結網路,並考慮點與邊的故障問題。當一個環能夠嵌入到一個網路中時,也就是相對映的圖形中包含有漢米爾頓迴圈,更進一步來說,當一個網路中壞掉k 個元件之後,還能嵌入一個環的話,那就是這個圖形是 k-漢米爾頓圖。有些研究人員曾研究過將容錯環嵌入到超立方體(Hypercube)與 star 網路,在本計畫中,我們將焦點集中在研究多容錯環狀網路嵌入到一些其他之重要連結網路上,例如:pancake 網路,k-ary n-cube網路 。 
英文摘要: An interconnection network connects the processors of the parallel computer. A ring network is one of the most fundamental topology for interconnection networks. A ring network is often used as a connection structure for local area networks, for example Token Rings. The graph embedding problem is one of the central issues in evaluating an interconnection network. Given a guest graph G and a host graph H, this problem is to find a mapping from every node of G to one of H and a mapping from every edge of G to one path in H. In our study, we find a mapping from every node of G to one of H and a mapping from every edge of G to one edge in H. Fault tolerance is also an important issue especially when the size of an interconnection network is large. In this project, we study the fault- tolerant ring network embedding in some important interconnection networks with both nodes and links failures. A graph can be used to represent an interconnection network where the nodes correspond to the processors and the edges to the communication links. Hence, we use graph and network interchangeably. Let the graph G=(V,E) represent an underlying interconnection network. Two types of failures in a multiprocessor system are of interest, processor failures and link failures. When a ring can be embedded a network, thus the graph contains a Hamiltonian cycle. Furthermore, when a graph is a k- Hamiltonian graph, the corresponding network is k-fault-tolerant for rings where the number of the faulty set F is k. Some researchers studied the fault-tolerant ring embedding in hypercubes, star- networks. In this project, we aim at studying k-fault-tolerant ring embedding in some important interconnection networks, such as pancake networks, k-ary n-cbues networks, and butterfly networks. 


科資中心編號: RB9009-0056 
題名: 星狀結構網路的容錯繞路方法研究 
Adaptive Fault-Tolerant Routing Algorithms for Star Networks 
研究者: 施釗德 ; 黃天佑 
Shih, Jau-Der ; Huang, Tien-Yu 
機構名稱: 屏東師範學院數理教育系 
經費來源: 行政院國家科學委員會 
補助金額: 新台幣 489 仟元 
補助編號: NSC89-2213-E153-001 
研究日期: 1999/08 - 2000/07 
出版日期: 2001/04 
頁冊數: 5頁 
原件提供: 行政院國家科學委員會科學技術資料中心, NSC89-2213-E153-001 
科資分類號: IB0203 
關鍵詞: 容錯繞線 ;星狀結構 ;分散式系統 ;平行處理 
Fault Tolerance Routing ;Star Interconnection Network ;Distributed System ;Parallel Processing 
中文摘要: 本研究計劃介紹一個在星狀(Star)多處理機系統架構中無死結且能容錯的繞路方法。當訊息在傳送的過程中並無遭遇到故障或不安全處理機時,訊息將以最短路徑送達目的地。若訊息在傳送過程中遭到故障的處理機所阻斷時,訊息將繞過這個故障的處理機而往目的地前進,在並非所有處理機皆為不安全的狀態下,訊息將以不超過2n的路徑送達目的地,其中n為星狀網路之維數。以往文獻上的結?G路徑皆為故障的處理機數目之函數,而本研究結果路徑與故障的處理機數目無關,且每個處理機只需紀錄鄰近處理機的狀態。 
英文摘要: This report presents an adaptive fault-tolerant deadlock-free routing algorithm for star networks. The algorithm can route a message via the shortest path as long as the path doesn't include any faulty or unsafe nodes. When the message is blocked by faulty nodes, the message will find a detour to route around the faulty nodes. Our algorithm can route a message via a path of length no more than 2n as long as not all nodes are unsafe or faulty, where n is the dimension of the star network. The length of the routing path is independent of the number of faulty nodes. 


科資中心編號: RS9010-0033 
題名: 在受損的超維立方體多處理機上設計一個高效率容錯之重組策略
Design of an Effectiviely Fault-tolerant Reconfiguration Scheme in Injured Hypercubes 
研究者: 許健平 ; 王志明 ; 張志浩 ; 劉安琦 ; 蘇元祺 
Sheu, Jang-Ping ; Wang, Chih-Ming 
機構名稱: 中央大學資訊工程系 
經費來源: 行政院國家科學委員會 
補助金額: 新台幣 349 仟元 
補助編號: NSC85-2213-E008-030 
研究日期: 1995/08 - 1996/07 
出版日期: 1996/07 
頁冊數: 35頁 
原件提供: 行政院國家科學委員會科學技術資料中心, NSC85-2213-E008-030 
科資分類號: IA0301 
關鍵詞: 容錯 ;超立方體 ;重組方法 ;矩陣相乘 ;排序演算法 
Fault Tolerance ;Hypercube ;Reconfiguration Scheme ;Matrix Multiplication ;Sorting Algorithm 
中文摘要: 在此計畫中,我們在受損的超維立方體多處理機上設計一個高效率的容錯重組策略,此容錯重組策略稱之為最大無錯子超維立方體---環狀架構。我們的容錯重組策略是由在受損的超維立方體中辨視出最大無錯子超維立方體所連成的環,且相鄰的子超維立方體之間的距離最多是3。最大無錯子超維立方體的建構是藉由辨視出最多數量的無錯子超維立方體,以達到高處理器使用率與高系統效能的目的。使用此容錯策略,我們可以在n維的超維立方體上容許超過n個錯誤並且仍能維持合理的高系統效率。為了論證此策略的容錯能力,我們在實際的平行機器nCUB E/2E含32個處理器上,實際製作兩個應用的演算法,矩陣相乘和排序演算法。實驗數據顯示出我們的策略確實具有低的系統效率降低率和高的處理器使用率。 
英文摘要: In this report, we present a reconfiguration approach to identify the maximal fault-free subcube-ring for tolerating faults in injured hypercubes. The fault-free subcube-ring is connected by a ring of fault-free subcubes with dilation 3. By exploiting the size of fault-free subcubes as large as possible, the maximal fault-free subcube-ring with higher processor utilization is obtained. Using this approach, we can tolerate more than n faults in n-dimensional hypercubes. To demonstrate the fault-tolerant capability of our approach, we implement two fault-tolerant algorithms, matrix- multiplication and sorting algorithms, on the nCUBE/2E hypercube machine with 32 processors. The simulation results show that our reconfiguration approach has low performance slowdown and high processor utilization. 


科資中心編號: RS9010-0011 
題名: 不規則聯結網路之研究 
A Study of Irregular Interconnection Networks 
研究者: 劉振緒 
Liu, Jen-Shiuh 
機構名稱: 逢甲大學資訊工程系 
經費來源: 行政院國家科學委員會 
補助金額: 新台幣 220 仟元 
補助編號: NSC84-2213-E035-005 
研究日期: 1994/08 - 1995/07 
出版日期: 1995/07 
頁冊數: 28頁 
原件提供: 行政院國家科學委員會科學技術資料中心, NSC84-2213-E035-005 
科資分類號: IA0202 
關鍵詞: 不規則連結網路 ;超立方體 ;容錯 ;一般化費氏立方體 ;資料通訊 ;圖形映埋 
Irregular Interconnection Network ;Hypercube ;Fault Tolerance ;Generalized Fibonacci Cube ;Data Communication ;Graph Embedding 
中文摘要: A graph is regular if each node has the same degree. Most interconnection networks are based on regular graphs. Very limited knowledge has been accumulated for irregular interconnection networks for lack of appropriates tools. Some results for a family of irregular interconnection networks, which consists of the generalized Fibonacci cube and the Incomplete Hypercubes, have been reported recently. This family of interconnection networks has two major areas of application: (1) Fault Tolerance - each member network serves as an alternative structure for reconfiguring a hypercube in the presence of faults, and (2) Incremental Expansion - allowing a system to add nodes in small increments, which is an important advantage for systems that evolve with time. In this research project we have studied a family of irregular interconnection networks. In particular, we present data communication primitives for the Incomplete Hypercubes and design some application algorithms for members of the family. 



科資中心編號: RB9012-0461 
題名: 網格連結網路上完全二元樹容錯嵌入之研究 
On the Fault-tolerant Complete Binary Tree Embedding of the Mesh Interconnection Network 
研究者: 徐俊傑 
Hsu, Chiun-Chieh 
機構名稱: 台灣科技大學資訊管理系 
經費來源: 行政院國家科學委員會 
補助金額: 新台幣 511 仟元 
補助編號: NSC89-2213-E011-129 
研究日期: 2000/08 - 2001/07 
出版日期: 2001/07 
頁冊數: 5頁 
原件提供: 行政院國家科學委員會科學技術資料中心, NSC89-2213-E011-129 
科資分類號: CA0102;IA0202 
關鍵詞: 網狀連接系統 ;完全二元樹 ;嵌入 ;容錯 
Mesh Connected System ;Complete Binary Tree ;Embedding ;Fault Tolerance 
中文摘要: 本計畫研究網格連結網路上在多維連結與光連結情況下的容錯嵌入問題。研究的方向在於提出有效方法以提高完全二元樹嵌入法在網格連結網路上的處理器使用效能,並使該嵌入結果具備容錯機制以保證結果之可靠度。利用網格連結網路所具有規則且可分解特性,為求達到理想的嵌入結果目標,我們以標記法配合遞迴法,在不同的多處理器連結網路上,首先建構一棵近似最大的最終完全二元樹,於此同時,我們也探討在其它未使用之閒置處理器上,構建其它的完全二元樹,或者是將這些閒置節點安排做為最終完全二元樹節點之替代節點,以解決樹節點發生障礙時之容錯與復原問題。在維持相同的條件限制之下,我們所提出的方法獲致了相對於先前其他相關研究較佳的嵌入結果,在三維的網狀圖上我們則進一步得到了最佳的嵌入結果。 
英文摘要: This project aims at resolving the problem of fault-tolerant embedding in the mesh interconnection network under multi-dimensional and optical connection. Since the processor is the most important resource in the multi-processor system, we focus on increasing the node utilization and providing the fault-tolerant mechanism at the same time when embedding a complete binary tree (CBT). With the capability of fault-tolerance, the embedding results become more guaranteed. By using the wormhole routing and optical transmission, we release the dilation constraint of tree edge and make the solution more flexible. By using decomposition properties of the mesh, we apply labeling scheme with recursive construction in order to achieve high processor utilization. First, a complete binary tree is constructed, and then we use those free nodes to construct other CBTs. Meanwhile, we treat all those free nodes as possible replacement nodes for faulty nodes. As a consequence, we obtain better results compared with previous studies. Furthermore, for the 3D mesh, the embedded CBT is of the largest size. 


科資中心編號: RB8910-0191 
題名: 開發一個遵循CORBA 2.0標準之容錯物件服務
Fault-Tolerance Object Service on CORBA 2.0 
研究者: 梁德容 
Liang, De-Ron 
機構名稱: 中央研究院資訊科學研究所(ASII) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 591 仟元 
補助編號: NSC88-2213-E001-022 
研究日期: 1998/08 - 1999/07 
原件提供: 行政院國家科學委員會科學技術資料中心 NSC88-2213-E001-022 
科資分類號: IA0301;IA0202 
關鍵詞: 共通物件策託經紀人架構 ;容錯物件服務 ;容錯 ;軟體可靠度 ;物件導向程式設計 
Common Object Request Broker Architecture (Corba) ;Fault Tolerance Object Service ;Fault Tolerance ;Software Reliability ;Object Oriented Programming (Oop) 
中文摘要: CORBA(Common Object Request Broker Architecture)在分散應用上已經漸漸普遍。它雖然簡化了分散系統的設計,但在CORBA的整個架構中,仍未提供完整的容錯服務。因此系統開發者必須自行發展容錯機制。設計容錯機制通常很煩人且易生錯誤。因此我們很需要有可複用及彈性的CORBA容錯機制。我們在本計畫中提出一個容錯的物件服務,稱之為FTOS( Fault-Tolerance Object Service)。應用程式利用FTOS 便可容忍像是Host crashes、Object crashes、Object hangs的錯誤。FTOS允許程設師選擇偵錯、復原、多重方式及可靠度以符合其需求。 
英文摘要: The use of Common Object Request Broker Architecture (CORBA) to implement distributed applications is becoming increasingly popular. Although CORBA simplifies the design and implementation of distributed applications, it does not facilitate the development of reliable distributed applications. As a result, application developers have to implement their own fault tolerant mechanisms. The task of implementing fault tolerant mechanisms is often tedious and error-prone. Therefore, there is a great need to make fault tolerant mechanisms in CORBA reusable and flexible. In this project, we propose a fault-tolerance object mechanism designed as a common object service, namely Fault-tolerant Object Service (FTOS). Through the use of FTOS, application object services can embed fault- tolerance support for tolerating failures such as host crashes, object crashes, and object hangs. FTOS allows application developers to select detection methods, recovery strategies, replication schemes and the degrees of reliability that are most suitable to their applications. 


科資中心編號: RB8904-0108 
題名: 具一般性的多計算機系統容錯模式
An Universal Fault-Tolerant Model on Multicomputers 
研究者: 楊竹星 ; 吳順裕 ; 朱榕屏 ; 蔡文山 ; 杜彥宏 
Yang, Chu-Sing ; Wu, Shun-Yue ; Zu, Long-Ping ; Tsai, Wen-Shan ; Tu, Yen-Hung 
機構名稱: 中山大學電機工程系(NCSEELE) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 361 仟元 
補助編號: NSC82-0408-E110-011 
研究日期: 1992/08 - 1993/07 
頁冊數: 35頁 
原件提供: 行政院國家科學委員會科學技術資料中心 NSC82-0408-E110-011 
科資分類號: IE0506;IA0201;IA0202 
關鍵詞: 容錯 ;連結網路 ;可靠度 ;備用點使用率 ;拓樸 ;硬體成本負擔 ;多電腦系統 
Fault Tolerance ;Interconnection Networks ;Reliability ;Spare Utilization ;Topology ;Overhead ;Multicomputer System 
英文摘要: An universal fault-tolerant model that can be applied in multiprocessors in proposed. The fault-tolerant model uses hardware redundancy to reconfigure a faulty system so that the topology with the original dimension can be retained. The switch implementation is investigated to reduce node degree. The result shows that the node degree is maintained for any fault combination. Generally, the model is suitable for every regular interconnection network. Furthermore, the fully spare utilization is achieved in each module, i.e., each spare node can be used to replace any faulty node within a module. Since the hardware overhead and system reliability are two main factors for measuring such type of systems, our scheme will be compared with other works in terms of the two factors. The result can show our scheme requires less hardware overhead and/or achieve higher system reliability. 


科資中心編號: RB8903-0002 
題名: 廣域性可重建容錯超晶格架構之研究 
The Study for a Reconfigurable Global Fault-Tolerant Hypercube Architecture 
研究者: 吳順裕 
Wu, Shun-Yue 
機構名稱: 陸軍軍官學校資訊工程系(CAAGINS) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 404 仟元 
補助編號: NSC88-2213-E145-003 
研究日期: 1998/08 - 1999/07 
頁冊數: 4頁 
原件提供: 行政院國家科學委員會科學技術資料中心 NSC88-2213-E145-003 
科資分類號: IA0101 
關鍵詞: 容錯 ;超立方體架構 ;可靠度 ;重建 ;切換網路 
Fault Tolerance ;Hypercube Architecture ;Reliability ;Reconfiguration ;Switching Network 
中文摘要: 本計畫擬提出一個廣域性容錯超晶格架構FT(N,k),此容錯架構FT(N,k),藉由連接N=2"個作用處理器與k個備用處理器,使用(n-1)個切換網路SN(N,k)完成超晶格架構緯度1至緯度(n-1)的連接,而緯度0則由處理器間線性陣列連接完成,而且使用切換器以便繞過故障處理器。每一個切換網路可由l階的 Decoupling networks來實現,其中L=[log/sub 2/(k+1)],且每一階的 Decoupling networks充分利用超晶格的連接特性來建立,如此可大幅降低硬體成本。若與相關論文比較時,本架構僅使用較少的硬體成本便可達成較高或同等級的系統可靠度,此外,本架構也能容忍切換器與連接線的錯誤。 
英文摘要: In this paper, we present a fault-tolerant hypercube architecture using switching networks to tolerate node, link and switch failures. An n-dimensional fault-tolerant hypercube consists of N (=2") active nodes (processors) and k spare nodes. The N+k nodes are connected by links in a ring structure to form the inter-connection of dimension 0 of the hyper-cubes. A switching network is used to implement interconnection along distinct dimension. Thus, we require n-1 switching networks and a ring structure to build the proposed fault-tolerant architecture. The proposed scheme compared with related works can achieve higher or the same level reliability using few hardware costs. In addition, The scheme can tolerate switch and link failures. 


科資中心編號: RB8805-0059 
題名: 模組分享容錯多處理機設計 
The Design of Module-Sharing Scheme for Fault-Tolerant Multiprocessors 
研究者: 吳順裕 
Wu, Shun-Yue 
機構名稱: 陸軍軍官學校資訊中心(CAAGCOC) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 303 仟元 
補助編號: NSC87-2213-E145-002 
研究日期: 1997/08 - 1998/07 
頁冊數: 4頁 
原件提供: 行政院國家科學委員會科學技術資料中心 NSC87-2213-E145-002 
科資分類號: IA0301;IA0203 
關鍵詞: 容錯 ;重建策略 ;可靠度 ;模組分享 ;多處理機 
Fault Tolerance ;Reconfiguration Strategy ;Reliability ;Module Sharing ;Multiprocessor 
中文摘要: 本計畫乃提出不限拓樸結構且具完全備用點使用率(Full spare utilization)的容錯多處理機架構,計畫中使用模組間備用點的分享達成完全備用點取代而非一般廣域性備用點分配(Global sparing)。本架構的建立,首先將所有原始點(Original nodes)及備用點(Spare nodes)分成數個模組(每個模組點數不限),先建立各個模組,再連接所有模組來建立完整系統,於任一模組內任何備用點皆可直接取代模組內或遠方模組的故障點,因此本架構具有廣域性備用點分配能力且能達到最大備用點使用率。為達此需求,我們發展一個重建演算法來分配備用點的使用,並且撰寫模擬程式模擬其結果,結果中顯示,相較於其他相當架構,本架構的重建效率相當高且只需較少硬體成本便可達到相同或較高的系統可靠度。 
英文摘要: This project presents a module-sharing scheme for constructing global fault-tolerant multiprocessors which, not limited to special topologies, can apply to any general multiprocessor systems. The system is constructed from modules in which any spare node in a module can be used to back-up any node failure in the same module or in any other remote module. Therefore, the proposed scheme possesses global sharing capability and achieves maximum utilization of spares. To meet such requirements, a reconfiguration strategy is developed to replace faulty nodes by allocating available spares (a simulation result will be appeared). The findings demonstrate that the efficiency of reconfiguration is superior to any other related works. Compared to certain other global or modular schemes, our scheme can achieve a higher level of reliability with less hardware cost. 


科資中心編號: RB8707-0045 
題名: 不停機分散式執行環境之設計與製作---子計畫(I):分散式系統錯誤診斷程序之設計與製作 
Design and Implementation of Diagnosis on the Distributed System 
研究者: 朱榕屏 
Zu, Long-Ping 
機構名稱: 永達工商專校電機工程科(YTICELE) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 302 仟元 
補助編號: NSC85-2213-E132-001 
研究日期: 1995/08 - 1996/07 
頁冊數: 28頁 
原件提供: 行政院國家科學委員會科學技術資料中心 NSC85-2213-E132-001 
科資分類號: IA0203 
關鍵詞: 容錯 ;錯誤診斷程序 ;自我測試 ;路由器 ;分散式系統 
Fault Tolerance ;Fault Diagnosis ;Self Testing ;Router ;Distributed System 
中文摘要: 此計畫將致力於診察判斷多重處理機系統內部的路由器錯誤。首先,我們介紹一個功能取向的路由器模型。根據此模型,路由器可被想像由一群連結點組合而成,每一個連結點負責該路由器某一組輸入輸出傳輸線的連結。因此,若路由器內部的某一條連結被發覺為故障,則與該連結有關的連結點而非整個路由器將被視為故障,於是此路由器內部的其他連結仍然能夠被繼續使用。此計畫將設計一個兩階段測試方法,用於測試多重處理機系統內部所有的連結點。測試的結果將交由錯誤診斷程序確認系統內部所有連結點的正確狀態。此計畫的貢獻說明如下。經由此計畫的路由器診斷程序,多重處理機系統內部所有的連結點狀態將被確認。因此,系統內部任何一個處理機將與其相鄰的處理機(藉由直接連結傳輸線)以及其他的處理機(藉由診斷為非故障的連結傳輸線)共同完成處理機的錯誤測試工作。因為參與共同測試的處理機數目確實大幅度增加,因此我們相信多重處理機系統的下一階段處理機錯誤診斷工作將因本計畫的成果受益許多。 
英文摘要: In this plane, the fault-diagnosis on the routers of multicomputer system is investigated. We first describe a connection- oriented router model. In this model, a router can be regarded as a matrix of crosspoints. Each crosspoint controls the connection of a pair of input and output links of the router. Thus, if a connection in a router is detected failed, then it is not the router but the crosspoints included in the connection to be considered failure. At this time, the other connection functions in that router can still be used. A two-phases testing method is designed to test the whole crosspoints in every router of multicomputer system. The test results will be used by our developed diagnosis algorithm to find which crosspoints failed. The contribution of the work in this plane is described below. After the diagnosis of routers, every crosspoint state will be identified. At this time, each node of system can take the neighboring nodes through the direct connections and many another nodes through the diagnosed nonfaulty connections (which are composed of only diagnosed nonfaulty crosspoints) as the testers. Since the number of testers for each node is significantely increased, the diagnosability of multicomputer system will be greatly improved. 

科資中心編號: RA8601-2041 
題名: 超立方體之高效率容錯策略
An Efficient Strategy for Tolerating Faults in Hypercubes. 
研究者: 陳省隆 
Chen,Hsing-Lung 
機構名稱: 台灣工業技術學院電子工程技術系(NTITENT) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 463 仟元 
補助編號: NSC84-2213-E011-014 
研究日期: 1994/08 - 1995/07 
出版日期: 1995/10 
頁冊數: 69頁 
原件提供: 行政院國家科學委員會科學技術資料中心 NSC84-2213-E011-014 
科資分類號: IA0301 
關鍵詞: 超立方體計算機 ;容錯 ;搜尋次立方體法 ;次立方體分割法 
Hypercube Computer ;Fault Tolerance ;Subcube Identification ;Subcube Partition 
中文摘要: 以往的超立方體容錯策略中,搜尋次立方體法(Subcube identification)不論故障節點數目多寡均可採用,但有系統效率(Performance)隨故障節點數目增加而快速下降的缺點;次立方體分割法(Subcube partition)則提供了一個系統效率下降較緩和的容錯策略,但此法只提出了在故障節點數目小於系統維度n時的容錯方法,當故障節點數目大於等於n時便沒有具體可行的方法。本研究計畫提出一結合搜尋次立方體法與次立方體分割法兩者方法與優點的超立方體高效率容錯策略,並且為此一新容錯策略發展一僅使用二層虛擬通道的蟲孔式容錯遞送演算法與廣播演算法。由模擬的結果證實本高效率容錯策略可以提供較佳的節點利用率,保持較高的系統效率。 
英文摘要: Fault tolerant strategies for hypercube computers proposed before can be divided into two classes: subcube identification and subcube partition. Subcube identification method can be applied on an faulty hypercube regardless of the number of faulty nodes; however, the size of recognized fault-free subcubes drops drastically. Subcube partition method just cut down the computation speed of the system by half; but it has practical implementation only when the number of faulty nodes is less than the number of dimensions of the hypercube. In this project, we present an efficient fault tolerant strategy which combines both advantages of subcube identification and subcube partition methods. This efficient fault tolerant strategy do not limit the number of faulty nodes and its performance degradation after injured is less than that of subcube identification method. In addition, deadlock free routing algorithm and broadcasting algorithm, both use only two layers of virtual channel, are proposed. The simulation results reveal that the efficient fault tolerant strategy has better fault tolerating ability than those of both subcube identification and subcube partition methods. 


科資中心編號: RA8510-2109 
題名: 順向類神經網路的容錯演算法與硬體備份研究 
Fault-Tolerant Algorithms and Hardware Redundancy of Feedforward Neural Network. 
研究者: 徐演政 
Hsu,Yen-Tseng 
機構名稱: 台灣工業技術學院電機工程技術系(NTITEET) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 201 仟元 
補助編號: NSC84-2213-E011-027 
研究日期: 1994/08 - 1995/07 
出版日期: 1995/07 
頁冊數: 136頁 
原件提供: 行政院國家科學委員會科學技術資料中心 NSC84-2213-E011-027 
科資分類號: IA0407 
關鍵詞: 神經網路 ;容錯 ;硬體備份 ;訓練演算法 
Neural Network ;Fault Tolerance ;Hardware Redundancy ;Training Algorithm 
中文摘要: 類神經網路因具有相當多的連結及神經元,故一般被認為具有內在的容錯特性。但因傳統的逆傳遞訓練演算法並沒有把加強類神經網路的容錯特性這項要求考慮進去,所以使用傳統的逆傳遞訓練演算法訓練完後的類神經網路即使有多餘的神經元其容錯能力也不會有顯著的提升。在本研究中,我們提出一個容錯訓練演算法,稱為分裂容錯訓練演算法,其可使神經網路在任何單一的隱藏層神經元壞掉的情況下皆和完整的神經網路的輸出相近似,我們稱此神經網路為有良好的容忍單一的隱藏層神經元壞掉的能力。且使用我們的分裂容錯訓練演算法並不須要花時間去找尋合適的隱藏層神經元個數,只要是在傳統的逆傳遞訓練演算法中可使用的隱藏層神經元個數皆可被使用來當作我們分裂容錯訓練演算法的初始隱藏層神經元個數。因為我們的分裂容錯訓練演算法會自然增長隱藏層神經元個數至適當的程度。經過我們的分裂容錯訓練演算法訓練之後的類神經網路壞掉一個隱藏層神經元時,雖然可正常的工作,但此時神經網路若再度壞掉一個隱藏層神經元則我們就不能保證神經網路還有良好的輸出。故我們提出一個再訓練的容錯演算法。此演算法被使用在經過我們的分裂容錯訓練演算法訓練之後的類神經網路壞掉一個隱藏層神經元時。使用此再訓練的容錯演算法可使神經網路再度保有容忍一個隱藏層神經元壞掉的能力。最後,我們討論三個故障模型,即輸入樣本有雜訊、連結斷掉、隱藏層神經元壞掉等。我們進一步地將我們的分裂容錯訓練演算法和其他可增加神經網路容錯能力的訓練演算法加以組合,稱為合成容錯訓練演算法,並且分析被合成容錯訓練演算法訓練之後的神經網路的容忍各種故障能力是否有增加。我們發現被某些合成容錯訓練演算法訓練之後的神經網路對某些故障模型的容忍能力確實有增加。此外我們亦研究用硬體備份方式以增加順向類神經網路之容錯能力,因此提出簡化的硬體備份的容錯順向類神經網路,爾後我們可整合容錯演算法及硬體備份兩種方法,使順向類神經網路之容錯能力得以更加提昇。 
英文摘要: Artificial neural networks have many links and nodes, so they have inherent fault-tolerant capabilities. But the traditional backpropagation (BP) training algorithm can't improve the fault- tolerant capabilities of neural networks significantly, even they have redundant nodes. In this research, we propose a training algorithm, called SFT (Split Fault-Tolerant) training algorithm that can improve the fault-tolerant capabilities of the neural networks effectively. The fault-tolerance of the neural networks means that they have almost equivalent output by removing any single hidden- layer node of neural networks. Also, the neural networks trained by SFT will increase hidden-layer node number naturally. Therefore, it does not need to find the suitable node number of hidden-layer. After removing one hidden-layer node of neural networks trained by SFT, it can also work normally, but it will not be good when one hidden-layer node is removed once again. Therefore, we propose a retraining fault-tolerant algorithm. This retraining fault- tolerant algorithm is used to training neural networks that removed one hidden-layer node and add a new hidden-layer node. It will let the neural networks have the fault-tolerant capabilities once again. Finally, we discuss three fault models: input patterns have noises, links are disconnected, and hidden-layer nodes are removed. We combine our proposed SFT training algorithm and other fault-tolerant training algorithms, and then we compare the fault-tolerant properties of neural networks trained by these hybrid training algorithms for these fault models. We find that their fault-tolerant properties are very good for some fault models. Finally, We also proposed a feedforward neural network with hardware redundancy to improve the fault-tolerant capability of neural network, Therefore, our research is a combination of fault- tolerant algorithms and hardware redundancy. It can increased the fault-tolerant capability of neural network. 


科資中心編號: RA8507-2091 
題名: 在不改變架構之前提下提昇類神經網路可靠度之研究 
The Strategy of Improving the Reliability of Neural Network without Changing the Network Structure. 
研究者: 謝文雄 
Hsieh,Wen-Shyong 
機構名稱: 中山大學資訊工程研究所(NCSUCIEG) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 337 仟元 
補助編號: NSC83-0408-E110-003 
研究日期: 1993/08 - 1994/07 
出版日期: 1994/11 
頁冊數: 68頁 
原件提供: 行政院國家科學委員會科學技術資料中心 NSC83-0408-E110-003 
科資分類號: IA0407 
關鍵詞: 類神經網路 ;容錯 ;可靠度 ;歸納 ;反向傳遞 ;多層感知網路 
Artificial Neural Network ;Fault Tolerance ;Reliability ;Generalization ;Backward Propagation ;Multilayer Perceptron 
中文摘要: 容錯是類神經網路中廣被強調的一個優點,然不論是在監督型或非監督型的訓練過程當中,吾人並不能控制此優良特性的走向,吾人能做的只是對一個已訓練完成的網路測試其究竟有多好的容錯能力。本研究旨在探討以逆向傳遞法則訓練之多層感知網路之容錯性質,並尋找一組鍵值使得當任何一個隱藏層節點發生故障時網路仍能提供可接受的效能。為達到此目的,在訓練的過程中除了使用降低均方根誤差之正規能量函數外,吾人另定義一個能量函數稱為限制型能量函數,其作用在於降低隱藏層節點故障時所產生的誤差。經由語音辨識和文字辨識兩個例子的驗證,吾人確能有效地找到一具有容錯能力的網路。除了容錯能力之外,經由正規和限制型能量函數訓練的網路亦具有較佳的歸納能力,之所以有如此的結果乃是限制型能量函數具有迫使輸出節點離開線性區進入飽和區的作用,因此,降低了節點和鍵值的靈敏度,本研究亦可擴展至容許多個節點故障。 
英文摘要: Fault tolerance capability is one of the frequently emphasized advantages of neural network. But during training, supervised or unsupervised, we can not control this good property, we only take this property as side effect of neural network. In this study, we exam the fault tolerant property of multi-layer perceptrons with one hidden layer which trained by back propagation. The study is focus on improving the fault tolerance capability and try to find out fault tolerant neural network which can tolerate single failure on hidden layer. This is done by forming additional energy term as constraint function and take this constraint function as part of training algorithm. Via speech and characters recognition two examples proved that we can find out a fault tolerant neural network efficiently. Experimental result indicate that network trained with constraint energy posses better generalization property than that trained by normal back propagation. The mechanism of constraint training proposed here has the tendency that forces the nodes and weights leaving out the linear region, thus reduce the sensitivity of weights and nodes. This study can be extended to tolerate multiple nodes failure. 


科資中心編號: RA8504-2038 
題名: 分散式系統有效復原演算法之研究 
A Study on Efficient Crash Recovery Algorithms in Distributed System. 
研究者: 莊東穎 
Juang,Tong-Ying 
機構名稱: 中華工學院資訊工程系(CHPICOS) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 114 仟元 
補助編號: NSC82-0113-E216-024-T 
研究日期: 1993/02 - 1994/01 
出版日期: 1994/01 
頁冊數: 29頁 
原件提供: 行政院國家科學委員會科學技術資料中心 NSC82-0113-E216-024-T 
科資分類號: IA0301 
關鍵詞: 容錯 ;分散式演算法 ;復原 
Fault Tolerance ;Distributed Algorithm ;Recovery 
中文摘要: 毀壞復原是分散式系統容錯計算中重要的一環。在分散式運算過程中,如果有一個或多個處理機損壞時,將造成系統的不一致。因此,如何使系統迅速而有效地恢復到一致的狀態,是毀壞復原所需要解決的重要問題。在過去的研究中,檢查點返回和訊息儲存是毀壞復原經常使用的方法。當損壞的處理機從最近的檢查點重新啟動後,正常的處理機經常會被要求返回到前面的某一檢查點或某一狀態。此種方法造成時間和計算資源的浪費。本計畫主要係根據樂觀訊息儲存法來儲存應用訊息。在系統正常運作時,當一個處理機收到的相鄰處理機送過來的應用訊息時,首先將送出此訊息的處理機的代號放到穩定記憶體中,然後將應用訊息以樂觀儲存法儲存起來。當系統中有處理機出問題時,可以利用該處理機中穩定記憶體裡過去所儲存的處理機的代號,使其恢復到發生損壞前的狀態。如此不但使系統快速地恢復正常,而且可以節省計算時間和資源,同時對於多個處理機毀壞的情況也能迅速而有效的加以解決。 
英文摘要: In distributed systems, capability to tolerate failures is an important and desirable property. Processor failures will lead the system to an inconsistent state. This proposal regards the research plan to obtain an efficient crash recovery algorithm which does not need the non-faulty processors to roll back in case of failure some other processors occur. Checkpointing, rollback and message logging have been widely used and studied by many researchers. In these methods, the faulty processor must restart execution from latest checkpoint while the non-faulty processors are usually needed to rollback to previous recovery point. Therefore a lot of overhead is incurred. In this report, application messages are logged optimistically after the sender identity of each application message is logged in a stable log by the receiver with this way, the original computation result of the application program on the failed processors can be restored such that time and computing resources are not wasted. In addition, it allows arbitrary number of processor failures. 


科資中心編號: RA8501-2025 
題名: 多層容錯性系統之運用
Applications of Multi-level Fault Tolerant System. 
研究者: 楊谷章 
Yang,Guu-Chang 
機構名稱: 中興大學電機工程系(NCHSELE) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 217 仟元 
補助編號: NSC83-0404-E005-002 
研究日期: 1993/08 - 1994/07 
出版日期: 1994/07 
頁冊數: 59頁 
原件提供: 行政院國家科學委員會科學技術資料中心 NSC83-0404-E005-002 
科資分類號: IA0201 
關鍵詞: 多階層容錯性系統 ;可靠度 ;暫態錯誤除刷 
Multi-Level Fault-Tolerant System ;Reliability ;Soft-Error Scrubbing 
中文摘要: 本計畫旨在探討多階層容錯性系統之可靠度。系統中之個別元件可以是一個小的細胞元或甚至一個複雜的微處理器。雖然這些元件個別的可靠度相當的高,但當大量的元件同時結合在單一大型系統中,系統的可靠度則降低。因此需要對多階層容錯性做整體的規劃分析,以建立較佳選擇容錯性階層的策略。本計畫的第一部分探討單階層容錯性系統之可靠度。系統組成的元件間彼此可以不相同,而且元件間可以有不同的錯誤速率,使系統分析將更適應於一般多樣性的系統。在本計畫的第二部分,則加入週期性暫態錯誤除刷技術對系統效能之探討。在本計畫的第三部分,則分析雙層容錯性系統效能並加入暫態除錯對系統影響之比較。以上分析的結果都均以實際系統模擬作驗證,為多階層容錯性系統作一完整之分析。 
英文摘要: This report is concerned with assessing the reliability of a two-dimensional array of components, in which each component may be anything from a binary cell---to a complex processor. Although individually these components are highly reliable, when a large number of them are combined in a single large system, the reliability of the system becomes worse. Therefore, a comprehensive framework to design and analyze such a fault-tolerant system has to be developed. In the first part of the report, we compute the reliability of a fault-tolerant system, which employs one level of fault-tolerance. The mean fault rates among the different components of the array can be different. Therefore, the result of the analysis can be applied to a wide variety of fault-tolerant systems. In the second part of the report, we extend previous work, which covers hard faults only, to include the technique of soft-fault scrubbing. In the third part of the report, the reliability of a system, which employs fault tolerance at two different hierarchical levels with soft-error scrubbing, is computed. Finally, the results are compared with a simulation of the precise system. 


科資中心編號: RA8410-2060 
題名: 分散式容錯伺服器之機制建立
The Design and Implementation of Mechanisms for Distributed Fault- Tolerance Server. 
研究者: 陳祈男 
Chen,Chyi-Nan 
機構名稱: 成功大學工程科學系(NCKEENS) 
經費來源: 交通部電信總局電信研究所(TCLG) 
補助金額: 新台幣 660 仟元 
補助編號: 83-企-第210號 
研究日期: 1993/08 - 1994/07 
出版日期: 1994/07 
頁冊數: 35頁 
科資分類號: IA0303 
關鍵詞: 容錯 ;可信賴伺服器 ;群體通訊 
Fault Tolerance ;Reliable Server ;Group Communication 
中文摘要: 分散式計算目前是計算技術中無可避免的趨勢,其中用戶/伺服器(Client/server)模式是一種非常適用於分散式計算的方法。而欲提高分散式系統的可靠度時,如何建立一具有容錯能力的可信賴伺服器(Reliable server)便成為一重要題目。本計畫的目的就是研究如何提供系統機制,以有效支援、建構具容錯能力之伺服器。在此計畫中,我們使用Group communication與伺服器群組模式(Server group model)來支援伺服器之容錯能力。我們將伺服器複製於一伺服器群組中以容忍計算機或網路的崩潰(Crash),伺服器群組可被視為一提供用戶容錯的單一個體。本文中先探討現有系統中ISIS與Amoeba的功能,再提出我們所制定的支援群組通訊(Group communication)之基元函數,以及所使用的通訊協定。最後舉出一實際應用範例,說明如何利用本系統去建立一個容錯伺服器。 
英文摘要: A distributed system has naturally the feature of hardware redundancy. How to increase the reliability of a distributed system becomes a software problem. In this paper, the feature of a fault tolerance system is studied and the primitive mechanisms based on the client/server model to support the building of a reliable server are provided. In our construction, the concept of group communication and server group model are used. A server should be replicated in different nodes; those servers are then executed in parallel, i.e., they work as a group, to provide the same service. Therefore, if there is a node crashed, the service can be continued by the other member of the group. To a user of the server, the service is always there, i.e., the server is reliable. In the paper, two systems, ISIS and Amoeba, are first studied; then the primitive functions are built; finally an example is given to demonstrate the use of those functions. 


科資中心編號: RA8404-2053 
題名: 星狀圖形連結網路之研究 
A Study on the Star Graph Interconnection Network. 
研究者: 楊昌彪 ; 韓慶智 
Yang,Chang-Biau ; Han,Ching-Chih 
機構名稱: 中山大學應用數學系(NSYCAPM) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 246 仟元 
補助編號: NSC83-0408-E110-010 
研究日期: 1993/08 - 1994/07 
出版日期: 1994/08 
頁冊數: 70 
科資分類號: IA;IA0201 
關鍵詞: 平行處理 ;星狀圖 ;廣播 ;排序 ;容錯 ;多處理機系 統 ;路程 
Parallel Processing ;Star Graph ;Broadcast ;Sorting ;Fault Tolerance ;Multiprocessor System ;Routing 
中文摘要: 多處理機系統的好處之一,就是它容錯的 能力。在本研究計畫中,我們將證明如果限制 此星狀圖形中任一節點的所有相鄰節點不可同 時發生故障,一個n維的星狀圖形其容錯能力是 2n-5。而且我們也將提出一個演算法來檢查有 2 n-4個節點發生故障的情況下,此n維星狀圖形是 否仍為連通圖。這個演算法需要O(n/sup 3/log n) 的時間。除此之外,我們將改進 Bagherzadeh等人所 提出之容錯資料繞徑(Routing)演算法,藉著快速 的計算一個排列的循環結構(Cycle structure)以及 在執行資料繞徑時,避免將資料送入堵絕節點 (Blocked node)。這些計算僅需常數的時間。然後 我們將提出一個有效率的容錯廣播(Broadcasting) 演算法。當沒有任何故障發生時,我們的廣播 演算法仍是最佳的演算法。而當最多不超過n-2 個節點發生故障時,額外所需時間是O(n/sup 2/)。星狀圖形上的排序問題已被學者解出,但 其時間複雜度尚未達最佳化,我們分析其演算 法,並提出一個在某些處理器或通訊連線故障 的情況下,改善其效率的方法。 
英文摘要: The capability of fault tolerance is one of the advantages of multiprocessor systems. In this report, we shall prove the fault tolerance of star graphs is 2n-5 with restriction to the forbidden faulty sets. And we shall propose an algorithm for examining the connectivity of star graphs when 2n-4 faults exist. The algorithm requires O(n/sup 3/log n) time. Besides, we improve the fault- tolerant routing algorithm proposed by Bagherzadeh et al. by calculating the cycle structure of a permutation and the avoidance of routing message to a node without another nonfaulty neighbors. This calculation needs only constant time. And then, we shall propose an efficient broadcasting algorithm. When no fault occurs, our broadcasting algorithm remains optimal. And the penalty is O(n/sup 2/) if at most n-2 faults exist. Sorting on the star graph has been solved by embedding a mesh on the star graph and a sorting scheme, Shear-sort, is applied on the embedded mesh. However, their sorting algorithm on the star graph did not reach optimality. We shall analyze the key point why the algorithm is not optimal and investigate the behavior of the sorting algorithm. We shall also present a performance improving scheme when there are some links (processors) fail. 


科資中心編號: RA8301-2011 
題名: 高次立方體系統容錯繞線之研究 
A Study on Fault-Tolerant Routing in Cube-Connected Multiprocessors. 
研究者: 邱舉明 
Chiu,Ge-Ming 
機構名稱: 台灣工業技術學院電機工程技術系 (NTITEET) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 282 仟元 
補助編號: NSC82-0404-E011-101 
研究日期: 1992/08 - 1993/07 
出版日期: 1993/07 
頁冊數: 48 
科資分類號: IA;IA0101 
關鍵詞: 高次立方體 ;繞線 ;容錯理論 ;死鎖 
Cube-Connected ;Routing ;Fault Tolerance ;Deadlock 
中文摘要: 在本計畫中,我們從兩個不同方面探討高 次立方體的容錯繞線的策略與方法。由於二次 元高次立方體的結構日益受到重視,我們的研 究主要是針對此一結構進行。首先針對一個有 部分元件損壞的系統,我們提出一套有效的容 錯繞線策略,目標在於為訊息的傳遞找出一條 最短的路徑。在本系統中,假設每一節點(或計 算單元)只含有其區域資訊,亦即只知道其相鄰 節點的狀態。我們的繞線策略比先前的策略容忍更多的錯誤,繞線的效率也比以前的方法好 。其中死結現象的避免也一併考慮在此方法中 。特別是,我們提出一種只需要五個訊息資源 即可避免死結發生的方法。在本研究的第二部 分,我們探討了在有限資源的情形下,具有彈性 的繞線方法,最短路徑的繞線雖然很好,但往往 會引起擁塞和可觀的延遲。我們放鬆了先前的 繞線法則,允許非最短路徑來繞過損壞的元件 或擁塞的中間節點。無死結也是本處考慮的重 點,同時亦考量當系統資源增加時對繞線功能 的影響。 
英文摘要: In this report we study the fault-tolerant routing on hypercube systems in two different aspects. We first develop an efficient fault-tolerant routing strategy for a hypercube system with faulty components. It aims at finding minimum paths for messages communicated among the nodes with small computation time and storage space. Each node is assumed to contain only its local information, i.e. the state of its nearest neighbors. It is shown that our strategy is able to tolerate more faults than the previously proposed strategy. Specific routing algorithm to route packets in the case of some peculiar fault situations is presented. Deadlock freeness is also considered in our model. In particular, we present a method whereby only a constant of five classes of communication resources is required to guarantee the property of deadlock freeness. In the second part of the research, we study flexible routing strategy in a hypercube under the constraint of limited communication resources. Shortest-path routing may sometimes cause congestion and hence render a considerable latency experienced by a packet. We relax the set of routing criteria proposed earlier. Non-shortest paths are allowed to be used to bypass faulty components or congested intermediate nodes. The concern here is also the deadlock freeness. The effects of increasing system resources to the routing function is addressed as well. 


科資中心編號: RA81-2379 
題名: 容錯類神經網路之研究 
The Research of Fault Tolerant Neural Network. 
研究者: 謝文雄 
Hsieh,Wen-Shyong 
機構名稱: 中山大學電機工程系(NSYEELE) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 347 仟元 
補助編號: NSC81-0404-E110-102 
研究日期: 1991/08 - 1992/07 
出版日期: 1992/11 
頁冊數: 50 
科資分類號: IA;IA0202 
關鍵詞: 類神經網路 ;容錯 ;描述語言 ;模擬環境 ;複聯 
Neural Network ;Fault Tolerance ;Specification Language ;Simulation Environment ;Redundancy 
中文摘要: 本計畫主要的目的在探究以Multiprocessor和 利用VLSI的技術在製作類神經網路時的容錯技 術。我們嘗試在沒有加入外部的硬體控制線路 下,能夠即時偵測出錯誤的神經節點,並在不必 中斷原來的學習和分類動作下完成以備用節點 取代錯誤的節點之工作,以使該神經網路具備 了容許錯誤的能力。我們提供一個簡單的方法 來達成這樣的目的。我們以節點為單元,依神 經傳導的特性,利用編碼理論的原理,及建立於 節點與節點之間資料傳輸的通信協定,以檢測 出錯誤的神經細胞及聯接線的錯誤。然後再以 我們提供的取代方案,將存放於錯誤的神經細 胞內之加權值(Weights)轉移至備用的神經節點之 上,以繼續維持原來的動作。利用本論文所提供的方法,可以檢測出資 料傳輸的錯誤、聯接線的錯誤和節點的錯誤, 並以備用節點來取代之。我們以Turbo C來模擬 整個神經網路的環境,並採用以Perceptron的模式 和BPN(Back propagation network)的學習規則來證明這 個容錯技巧。 
英文摘要: In this report, we explore the fault-tolerant technique of the neural networks which are implemented by VLSI chips or multiprocessor. We try to find out a scheme that can detect the fault nodes and replace it with spare nodes on line. In VLSI implemented neural network, the code theory is used to detect the failue nodes and transmition links. In multiprocessor implemented neural network, the common bus is used as transmition media between layers, the communication protocol is used to detect the failue of network. A fault node will be found by the nodes in upper layer and lower layer, the respected parameters will be marked as failue and moved to the first spare node in the fault layer. The spare node will play the role of the fault node, so that the operation of network can be hold consistently. Based on the fault tolerant scheme, the failue in neural network can be detected and recoveried on line. A turbo C written program is designed to simulate the neural network environment. Perception and BPN (back propogation network) are used to confirm the fault tolerant scheme. 


科資中心編號: RA81-2166 
題名: 超立方體計算機之可容錯的資源管理 
Fault-Tolerant Resource Management in Hypercube Computers. 
研究者: 陳省隆 
Chen,Hsing-Lung 
機構名稱: 台灣工業技術學院電子工程技術系 (NTITELN) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 288 仟元 
補助編號: NSC81-0408-E011-04 
研究日期: 1991/08 - 1992/07 
出版日期: 1992/09 
頁冊數: 150 
科資分類號: IA;IA0201 
關鍵詞: 容錯 ;超立體計算機 ;資源佈置 ;系統效能 
Fault Tolerance ;Hypercube Computer ;Resource Placement ;System Performance 
中文摘要: 本報告描述我們所設計的演算法,它有效 且可容錯地管理超立方體計算機中的資源。雖 然超立方體計算機中的資源可由數個節點共用 以降低成本,但是多份的共享資源可減少只使 用一份共享資源所引起的潛在延遲。當一份資 源由於故障無法被使用時,導致使用其餘共享 資源的延遲增加,整個系統的效能可能因此降 低。假若適當地安置多份資源,使得在有故障 發生時,每個節點還能以相同的Hops數取得資源,則潛在的效能降低量可維持最小。本計畫提出 了一種資源存取策略,在有些資源由於故障無 法使用時,它能使資源的存取負荷依然保持均 衡,這樣能使系統的效能維持最好狀態。Rabin已提出一種方法將一個資料檔案F ,轉 換成n個編碼項(Encoded-pieces),對於一個已定的m 值,任意m個編碼項已足夠重建資料檔案F。我們 不複製原始資料檔案,而將多份的n個編碼項布 置於超立方體計算機中,可能不僅可節省儲存 空間又可容錯。假若適當地佈置多份的n個編 碼項,使得在有故障發生時,每個節點還能以相 同的Hops取得m個不同編碼項,則潛在的效能降低 量不致降低太多。 
英文摘要: The algorithms to efficiently and fault-tolerantly managing resources in hypercube computers are described. While a certain resource in the hypercube computer may be shared by several cube nodes to lower the cost, multiple copies of a shared resource often exist in the hypercube to reduce the potential delay in accessing any shared copy. After one resource copy becomes unavailable due to a fault , overall system performance could degrade due to an increase in the access delay of the remaining shared copies. Potential performance degradation is kept minimal if the resource copies are placed in a way that access to the resource is guaranteed to take the same number of hops from every node even in the presence of faults. The method of accessing resources is proposed to ensure that accessing load of resource copies are balanced after some resource copies become unavailable due to faults, which keeps system performance maximal. Rabin has proposed a method to translating a data file F into n encoded-pieces, any m of which suffice for reconstructing F for a given mn. The copies of n encoded-pieces, instead of duplicates of a original data file, may be placed in hypercube computers such that both efficiency in storage space and fault-tolerancy can be achieved. Potential performance degradation is also kept minimal if the copies of n encoded-pieces are placed in a way that access to any m different encoded-pieces is guaranteed to take the same number of hops from every node even in the presence of faults. 

科資中心編號: RN9101-0349 
題名: 使用模糊多準則測度消費者涉入 
研究者: 徐村和 
機構名稱: 義守大學企業管理系 
經費來源: 行政院國家科學委員會 
補助編號: NSC89-2416-H214-031-SSS 
研究日期: 2000/08 - 2001/07 
出版日期: 2001/07 
頁冊數: 3頁 
原件提供: 行政院國家科學委員會科學技術資料中心, NSC89-2416-H214-031-SSS 
科資分類號: IB1101 
關鍵詞: 消費者行為 ;產品涉入 ;模糊集合理論 ;模糊測度 ;整合運算子 
Consumer Behavior ;Product Involvement ;Fuzzy Set Theory ;Fuzzy Measure ;Aggregate Operator 
中文摘要: 在產品涉入的衡量上過去的研究,都將涉入以高低二分法歸類,但涉入應是一個連續性變數,因此將它刻意的二分會把兩端以外的中間層次強制歸入兩極內,導致中間層次的重要性被忽略,因此本研究嘗試運用模糊數學來衡量消費者產品涉入程度。以隸屬函數來表示涉入度的連續性變化,採用多準則評估方法測度消費者的產品涉入,由實證結果,顯示模糊測度模式,有相當的合理性及實用性。 
英文摘要: In this study we focus on the product involvement measurement. The involvement degree is often roughly divided into “High involvement” and “Low involvement”. However, the involvement degree is a continuous variable, not a “High” or a “Low” variable, In this way the human beings’ fuzzy cognition can’t be reflected. So in this study we use the fuzzy mathematics to measure the consumers’ product involvement degree. We use membership function to denote the continuous change in involvement. We also use multiple criteria evaluation method to measure the product involvement of the consumers. The results of the empirical study show that our model is more reseasonable and practicable. 


科資中心編號: RB9102-0829 
題名: 模糊集合經運算後其模糊度變化之研究 
The Study for the Fuzziness Variation of Fuzzy Sets Through Simple Calculation 
研究者: 邱智煇 ; 劉如熹 
Liu, Ru-Shi 
機構名稱: 清雲技術學院電子工程系(科) 
經費來源: 行政院國家科學委員會 
補助金額: 新台幣 155 仟元 
補助編號: NSC89-2218-E231-007 
研究日期: 2000/12 - 2001/07 
出版日期: 2001/10 
頁冊數: 8頁 
原件提供: 行政院國家科學委員會科學技術資料中心, NSC89-2218-E231-007 
科資分類號: IE0601 
關鍵詞: 模糊數 ;熵 ;模糊集合 ;模糊性 
Fuzzy Number ;Entropy ;Fuzzy Set ;Fuzziness 
中文摘要: 本計劃主要是探討幾個模糊集合經加減運算後其熵值變化的情形,同時提出方法來快速且有效的求得運算後之模糊集合的熵值。 
英文摘要: Entropy is one of the measure of fuzziness degree for a fuzzy set. This study is shown that the entropy of the resultant sum can be obtained easily from the sizes of all addend fuzzy numbers’ supports and any one addend fuzzy number’s entropy. Therefore, the entropy calculation in addition and subtraction becomes very simple and the calculation of the relative fuzzy number’s membership functions is avoided. 


科資中心編號: RN9101-0221 
題名: 以類神經網路驗證及評估模糊知識庫系統(II) 
Verification and Validation of Fuzzy Rule-Based Systems---A Neural Network Approach (II) 
研究者: 李漢銘 
機構名稱: 台灣科技大學電子工程系 
經費來源: 行政院國家科學委員會 
補助編號: NSC89-2218-E011-016 
研究日期: 2000/08 - 2001/07 
出版日期: 2001/07 
頁冊數: 7頁 
原件提供: 行政院國家科學委員會科學技術資料中心, NSC89-2218-E011-016 
科資分類號: IA0407 
關鍵詞: 模糊類神經網路 ;規則驗證 ;效能評估 ;驗證 ;規則庫系統 
Fuzzy Neural Network ;Rule Verification ;Perormance Evaluation ;Verification ;Ruled Base System 
中文摘要: 本年度計劃依據前一年之成果,進一步提出一個具有梯形模糊集合輸入之類神經網路知識庫系統KBNN/TFS ,此一架構能夠處理梯形模糊集合輸入,並且具有模糊規則修正、驗證以及模糊規則產生等特點。基於KBNN/TFS ,我們提出一個有效的方法來評估模糊推論系統的複雜度,此外,我們也提出三個簡化模糊類神經網路的方法來增進模糊推論的效率,第一個方法為模糊列表法(Fuzzy tabulation method),它可以被用來藉由修改模糊規則的前提來作模糊規則的結合,並且可以剔除一些在推論過程中無效的模糊變數;第二個方法稱為遞移模糊規則的方法(Transitive fuzzy rule compacting method),它主要是結合遞移關係(Transitive relation)來降低模糊推論的複雜度;第三個方法稱為相同前提之一致化的方法(Identical antecedent unifying method),它可以藉由替換具有單一相同前提之模糊規則方式來簡化多餘的模糊規則。利用以上這三種方法可以有效降低模糊規則結構的複度降,此外,本計劃也提出基於前面三個有效化簡模糊規則的評估驗證模糊規則方法,實驗結果顯示在經過這三個方法處理過後的模糊規則,其推論的效率可以被有效的提昇。本研究之部分成果發表於電機電子工程學會IEEE 所舉辦之第十屆Tools with Artificial Intelligence 國際研討會。 
英文摘要: Knowledge-Based Neural Network with Trapezoidal Fuzzy Set (KBNN/TFS) is a fuzzy neural network model, which handles trapezoidal fuzzy inputs with the abilities of fuzzy rule revision, verification and generation. Based on KBNN/TFS, an efficiency validation method is proposed to evaluate the rule inference complexity on KBNN/TFS. Besides, three methods that simplify the structure of this fuzzy rule -based neural network model are provided to enhance the inference efficiency. Fuzzy tabulation method, the first method, is performed to do rule combination by modifying the antecedents of some specific rules and then to eliminate the don't care variables in the rules. The second method, named transitive fuzzy rule compacting method, combines the rules with the transitive relations to decrease the computational load of inference. The third method, called identical antecedent unifying method, simplifies the redundant antecedents of rules by replacing the identical antecedents of the rules with a single specific antecedent. By these methods, the structure of rules can be simplified without changing the results of its inference. The proposed efficiency validation method is used to analyze and support the results of performing these three efficiency enhancing methods. Also the simulation results show that the efficiency is enhanced after performing these three efficiency enhancing methods. 


科資中心編號: RB9110-0005 
題名: 模糊派屈網路之專家系統建立及其在橋梁損傷評估之應用(II) 
Bridge Damage Assessment through a Fuzzy Petrinet Based Expert System(II) 
研究者: 蔣偉寧 
Chiang, Wei-Ling 
機構名稱: 中央大學土木工程研究所 
經費來源: 行政院國家科學委員會 
補助金額: 新台幣 273 仟元 
補助編號: NSC90-2211-E008-063 
研究日期: 2001/08 - 2002/07 
出版日期: 2002/10 
頁冊數: 6頁 
原件提供: 行政院國家科學委員會科學技術資料中心, NSC90-2211-E008-063 
科資分類號: IG0003 
關鍵詞: 模糊理論 ;派屈網路 ;橋梁損害評估 ;專家系統 ;不確定性 
Fuzzy Theory ;Petri Network ;Bridge Damage Assessment ;Expert System ;Uncertainty 
中文摘要: 本文提出一個新的模糊派翠網路與其演算法,並將其實作成混合式派翠網路系統-HPNS,以模擬我們為了同時處理專家系統內的不確定性與不精確性而提出的模糊法則式推理。此模糊法則式推理是一個融合可能性推理與模糊推理的綜合性推理方法。在本文的模糊派翠網路中,我們堅持原派翠網路的引發法則,那就是當一個Transition被引發之後,其所有輸入Place內的所有Token皆被移除,並有新的Token被適當的置於輸出Place內。此外,我們亦堅持原派翠網路內對 Marking的定義與語意。因此,原派翠網路已發展完善的分析方法,可檢視系統特性如Liveness、Safeness、Boundedness、Deadlock等,大多數皆可應用於本文的模糊派翠網路。本文提出的推理演算法不僅符合法則式推理的原則,原派翠網路的執行方式亦受到保持。 
英文摘要: There has been an increasing interest in developing bridge management systems (BMS) to alleviate the problems arising from old bridges. A key to success in BMS relies heavily on the reliability of the techniques adopted for damage assessment. It is widely recognized that the use of expert systems to capture the expertise and to mimic the reasoning patterns of experts on damage assessment is a promising direction towards BMS. However, several important issues need to be addressed in using expert systems for bridges damage assessment, such as: (1) linguistic assessment by experts usually takes the form of natural language that contains intrinsic imprecision, (2) uncertain and imprecise information involved in the damage assessment usually make the problem even harder, (3) the efficiency of fuzzy rule-based reasoning needs to be improved, and (4) an explanation facility is essential to increase the confidence about the assessment results. To address the issues discussed in the previous section, this research is aimed: (1) to develop a reasoning mechanism in an expert system, which is capable of dealing with uncertain and imprecise information. (2) to integrate expert systems with Petri nets. Petri nets with a powerful modeling and analysis ability are capable of providing a basis for variant purposes, such as knowledge representation, reasoning mechanism, knowledge acquisition, and knowledge verification. (3) to develop modularized rule bases for damage assessment of bridges and provide a fuzzy Petri net-based expert system tool in the Internet, based on the proposed methodology, that can assist the assessment of bridge damage. 


[科資中心編號]RB8910-0264
[題    名]
關於模糊資料的變異數分析
Analysis of Variance for Fuzzy Data
[研 究 者 ]
吳憲忠
Wu, Hsien-Chung
[機構名稱  ]暨南國際大學資訊管理系(CNUINM)
[經費來源  ]行政院國家科學委員會(NSC)
[補助金額  ]212000
[補助編號  ]NSC88-2218-E260-005
[研究開始日期]1998/12
[研究結束日期]1999/08
[語    文]中文
[建檔日期  ]2001/06
[出版情形  ]完成日期:民國88年12月
[原件提供  ]行政院國家科學委員會科學技術資料中心
NSC88-2218-E260-005
[科資分類號 ]IA0407; IA0401
[關 鍵 詞 ]
模糊資料; 變異數分析; 模糊集合; 數學規劃

Fuzzy data; Analysis of variance (ANOVA); Fuzzy set; Mathematical
programming

[摘    要]
我們提出了如何使用變異數分析來處理模糊資料的觀念和方法。為了能夠使用一般變異數分析的方法,我們將處理模糊資料的h-level set。因為模糊資料的h-level set為一般的實數值資料,因此一般變異數分析的方法將能夠使用。我們也提出了決策法則來接受或拒絕檢定假說。然而,為了得到Pessimistic degree和Optimistic degree,我們必須解四個最佳化問題。這兩種degree的觀念將被用來做接受或拒絕檢定假說。

The concept and method of imposing imprecise (fuzzy) data
upon the usual method of the analysis of variance (ANOVA) is
proposed. We transact the h-level set of fuzzy data for the sake of
invoking the usual method of analysis of variance for real-valued
data. We propose the decision rules about the acceptance and
rejection for testing hypotheses with the notion of the pessimistic
degree and optimistic degree by solving four mathematical
programming problems. Finally, we provide an example to clarify the
theoretical discussions.

[科資中心編號]RB8910-0276
[題    名]
利用模糊高階方法建構智慧型評量系統
Constructing an Intelligent Ranking System with Fuzzy Higher-Order
Structure
[研 究 者 ]
李麗華
Le, Lee-Haw
[機構名稱  ]朝陽科技大學資訊管理技術系(CYUTINA)
[經費來源  ]行政院國家科學委員會(NSC)
[補助金額  ]246000
[補助編號  ]NSC88-2416-H324-006
[研究開始日期]1998/08
[研究結束日期]1999/07
[語    文]中文
[建檔日期  ]2001/06
[出版情形  ]完成日期:民國88年12月
[原件提供  ]行政院國家科學委員會科學技術資料中心
NSC88-2416-H324-006
[科資分類號 ]IA0407
[關 鍵 詞 ]
智慧型模糊評量系統; 知識庫; 多準則; 決策支援系統; 模糊數; 模糊集合

Intelligent fuzzy ranking system (IFRS); Knowledge base; Multiple
criteria; Decision support system; Fuzzy number; Fuzzy set

[摘    要]
模糊排序方法被廣泛應用在各種模糊應用系統領域中,然而針對不同模糊數(集)特質必需採用不同的排序方法,這使得開發智慧型模糊應用系統有所不便,因為不論採用哪一種排序方法都未能適用於所有的模糊數(集)。本研究主要在建構一智慧型模糊評量系統(IFRS),此一系統可以針對不同之模糊數(集)自動找出適合的排序方法,這個機制是藉由本研究所建立的模糊排序知識庫來達成,除此之外,本研究亦加上Yager的高階架構(Second-order Structure),來解決排序後仍難分辨的情形。本研究所建構的IFRS系統己驗證可以被廣泛的應用在任何需用到評量方法的研究領域上。

Fuzzy ranking methods have been highly utilized in many areas
to help making decisions or to act like a defuzzification process
for interpreting the consequence. However, to better handle various
fuzzy numbers or fuzzy sets a proper fuzzy ranking method should be
carefully selected. This also implies one ranking method does not
guarantee to solve all various types of fuzzy sets. In consequence,
a fuzzy application system, if it takes various fuzzy numbers (or
sets) and uses only one ranking method, may have troubles to
generate the proper result. It is, hence, the goal of this research
to construct an Intelligent Fuzzy Ranking System (IFRS) to invoke
proper ranking method when meets various types of fuzzy numbers or
fuzzy sets. This research, first, studied 15 most commonly used
ranking methods and selected five ranking methods which were widely
used and covered wide spectrum of the application areas. The second
step of this research is to construct if-then rules which encodes
these five ranking methods in the ranking knowledge-base (RKB).
With RKB, the proper ranking method can be automatically inferred
based on the given input of fuzzy sets (or numbers). In addition
to RKB, Yager's second-order structure is incorporated in the
system. This process is to further differentiate fuzzy sets (or
numbers) when they are initially ranked similar but with different
shapes. The contribution of this research is to provide an
intelligent ranking system that can be applied or attached to any
fuzzy application, especially in the evaluation areas, so that the
searching time for ranking method can be eliminated and proper
ranking results can be efficiently obtained without spending extra
research time.

[科資中心編號]RB8910-0247
[題    名]
含有複合模糊屬性的模糊查詢之研究
Handling Fuzzy Queries that Contain Compound Fuzzy Attributes
[研 究 者 ]
王學亮
Wang, Shyue-Liang
[機構名稱  ]義守大學資訊管理系(ISUIFM)
[經費來源  ]行政院國家科學委員會(NSC)
[補助金額  ]321000
[補助編號  ]NSC88-2213-E214-006
[研究開始日期]1998/08
[研究結束日期]1999/07
[語    文]中文
[建檔日期  ]2001/06
[出版情形  ]完成日期:民國88年12月
[原件提供  ]行政院國家科學委員會科學技術資料中心
NSC88-2213-E214-006
[科資分類號 ]IA0407; IA0304
[關 鍵 詞 ]
模糊資料庫; 複合模糊屬性; 模糊查詢; 資料庫綱目; 模糊集合

Fuzzy database; Compound fuzzy attribute; Fuzzy query; Database
schema; Fuzzy set

[摘    要]
本研究之目的在於探討如何處理含有複合模糊屬性(Compound fuzzy attributes)之模糊查詢的理論與技術。所謂的複合模糊屬性是指原本並未定義於資料庫綱目(Schema)中,但可由綱目中原有的數個固定屬性合成而得之屬性。允許複合模糊屬性存在於查詢語句中並且具備處理此種屬性之資料庫系統,可減少查詢時得到虛解(Null answers)的情況,因此可使查詢更具彈性及人性化,從而提高系統之功能。針對由區間值、詞語(Scalar)、及其模糊集合所合成之複合模糊屬性,本研究工作提出一方法以處理具有此類複合模糊屬性之模糊查詢。我們的工作是[7]的延伸但卻涵蓋其所能處理之數字型複合模糊屬性。我們的研究成果則已分別發表於一篇期刊論文,四篇會議論文[9-11,15,16]。且正繼續研究探討並將陸續發表其結果[12-14]。

The goal of this project is to study how to handle fuzzy
queries that contain compound fuzzy attributes. Compound fuzzy
attributes are ambiguous attributes that are not defined in the
original database schema, but can be derived from multiple rigid
attributes in the schema. Database systems that permit these types
of attributes in the queries and provide capabilities to handle
them can certainly reduce the possibility of null answers.
Consequently, it makes user queries more flexible and
human-oriented, and system performance better improved. For
compound fuzzy attributes compounded from interval values, scalars
and their fuzzy sets, we have proposed a method to handle fuzzy
queries that contain these compound fuzzy attributes. Our work is
an extension of [7] but can also handle the compound fuzzy
attributes compounded from numbers. Our research results have been
published in one journal paper and four conference
proceedings [9-11, 15, 16] while continuing efforts will be
published shortly [12-14].


[科資中心編號]RF8801-0080
[題    名]
模糊資料包絡分析模式之求解與應用
Fuzzy Data Envelopment Analysis---Solution and Application
[研 究 者 ]
高強
Kao, Chiang
[機構名稱  ]成功大學工業管理科學系(NCKBIMS)
[經費來源  ]行政院國家科學委員會(NSC)
[補助金額  ]613600
[補助編號  ]NSC87-2416-H006-022
[研究開始日期]1997/08
[研究結束日期]1998/07
[頁 冊 數 ]58頁
[語    文]英文
[出版情形  ]完成日期:民國87年9月
[原件提供  ]行政院國家科學委員會科學技術資料中心
NSC87-2416-H006-022
[科資分類號 ]SM0405
[關 鍵 詞 ]
模糊集合; 資料包絡分析法; 分析模式; 生產力; 效率評估

Fuzzy set; Data envelopment analysis; Analysis model; Productivity;
Efficiency evaluation

[摘    要]
1978年Charnes、Cooper與Rhodes提出資料包絡分析模式後,即成為評估決策單位多投入及多產出效率之一有效方法,也廣泛應用於各領域。當資料包絡法中的投入或產出項為口語化述詞或模糊數值時,則模式即延伸為模糊資料包絡法。研究中利用.alpha.截集(.alpha.-cut)及擴展原則(Extension principle)的方法,將模糊資料包絡法,簡化成一對含有.alpha.水準參數的傳統資料包絡分析模式,並求解出效率值的下限及上限;藉由多個不同.alpha.值的效率值解,即可構建出效率值的隸屬函數。由於應用模糊資料包絡法所求得的效率值為模糊數值形式,因此很難直接依效率值的大小將受評單位排序。研究中應用「最大及最小集合」模糊數值排序的方法,來解決模糊效率值排序的問題。在實例應用方面,則以國內機械業之生產力、技術與管理指標等,在模糊情形下,評估各廠商之生產力效率,同時和前人在確定情形下之評估結果做一比較,以了解未採用模糊分析法之缺失。

Since its development in 1978 by Charnes, Cooper, and Rhodes
data envelopment analysis (DEA) has been widely applied to
different areas for evaluating the efficiency of multiple-input and
multiple-output decision making units (DMU). The existing DEA
models are limited to crisp data. This research proposes a
procedure to measure the efficiencies of DMUs with fuzzy
observations. The idea is based on the .alpha. cuts and extension
principle to reduce a fuzzy DEA model to a family of crisp DEA
models, which can be described by a pair of parametric programs, to
find the .alpha.-cuts of the membership functions of the efficiency
measures. Since the efficiencies calculated are also fuzzy numbers,
an appropriate ranking method is required to determine the order of
each DMU. In this research, the maximum set and minimum set method
is applied to secure correct ranking. To illustrate the validity of
the developed methods, the productivity, technology index, and
management index of Taiwan machinery firms are collected to
calculate the productivity efficiencies. The results are compared
with the previous results under crisp assumptions to see the
inappropriateness of not using fuzzy analysis.

[科資中心編號]RB8705-0244
[題    名]
模糊邏輯模式的最適化設計法則
The Design Algorithm for Optimum Fuzzy Set Logic System
[研 究 者 ]
郭林恩
Kuo, Lin-En
[機構名稱  ]元培醫專環境工程衛生科(YPMTEEH)
[經費來源  ]行政院國家科學委員會(NSC)
[補助金額  ]172000
[補助編號  ]NSC86-2212-E264-001
[研究開始日期]1996/08
[研究結束日期]1997/07
[頁 冊 數 ]5頁
[語    文]中文
[出版情形  ]完成日期:民國87年2月
[原件提供  ]行政院國家科學委員會科學技術資料中心
NSC86-2212-E264-001
[科資分類號 ]IB0205
[關 鍵 詞 ]
模糊集合論;模糊邏輯系統;歸屬函數;基因演算法;迴歸分析

Fuzzy set theory;Fuzzy logic system;Membership function;Genetic
algorithm;Regression analysis

[摘    要]
對程序控制、材料性質預測和程序最適化等應用而言,如何尋找一個準確的模式用以預測實際系統未來的發展是很重要的步驟。模糊邏輯系統(Fuzzy logic system,FLS)是近年來頗為盛行的一種黑箱模式的建構工具。本文採取Beta-M函數為FLS的歸屬函數目的在減少FLS所需的歸屬函數數目。在設計FLS最適化結構的過程中則是利用基因演化法則決定Beta-M函數的最佳參數值和FLS中模糊規則的最佳數目。結合模糊邏輯系統、基因演化法則的設計法則可以在可容忍的誤差內找出參數個數最少的模式。最後以一個電腦模擬的酸鹼反應槽結果印證此設計法。

It is an important step to develop an accurate process model to
predict the further system responses for many applications such as
process control, material property prediction, and optimization.
Fuzzy Logic System (FLS) is a popular black-box modeling technique in
recent years. In this work, the Beta-M function is adopted in order
to reduce the number of membership functions used in the FLS. During
the design stage, the Genetic Algorithm (GA) is used to find the
optimum parameters of Beta-M functions and the optimum number of
fuzzy rules used in the FLS. Combining the FLS and GA, a new FLS
design method is derived. With this design method, in a tolerable
error range, we can construct a FLS model with minimum number of
model parameters. A computer simulation of acid-base reaction is used
to show the effectiveness of this new FLS design method.

[科資中心編號]RB8701-0529
[題    名]
模糊數學於專案網路規劃之應用
A Study on Project Management-Application of Fuzzy Set Theory.
[研 究 者 ]
陳惠國
Chen, Huey-Kuo
[機構名稱  ]中央大學土木工程系(NCUECIE)
[經費來源  ]行政院國家科學委員會(NSC)
[補助金額  ]138600
[補助編號  ]NSC85-2211-E008-010
[研究開始日期]1995/08
[研究結束日期]1996/07
[頁 冊 數 ]51頁
[語    文]中文
[建檔日期  ]1998/10
[出版情形  ]完成日期:民國85年8月
[原件提供  ]行政院國家科學委員會科學技術資料中心
NSC85-2211-E008-010
[科資分類號 ]IG0406
[關 鍵 詞 ]
模糊理論;多評準決策分析;專案評核術;頂點法;要徑;擴展定理;專案管理

Fuzzy theory;MCDM;PERT;Vertex method;Critical path;Extension
principle;Project management

[摘    要]
專案評核術(Program evaluation and review technique,PERT)已經被廣
泛的應用於多種領域,以提高專案的績效。有關專案網路的實際作業時間之估計,已使用過許多方法,其中以使用模糊集理論的估計方法較符合真實世界的合理性。然而,應用模糊運算於前進掃曳法及後退掃曳法中,以搜尋網路中的要徑,多種作業時間的出現是無法避免的。當使用擴展定理(Extension principle)時,將會造成時間區間(Time intervals)擴散的現象,此現象將造成搜尋要徑的困難,故此時間區間會擴張的現象必須消除。有效的使用專案資源是本研究的第二個研究主題。事實上,此問題在過去幾年已受到極大的重視,但有關的研究都著重於確定性的問題。由於自然資源使用上的不確定性,所以採用模糊數學模型來處理排程的問題,可以使專案達到在最短時間內完成及資源拉平的目的。在本研究中,將針對資源使用的不確定性以建構二個多目標、單一計畫且具選擇性作業模式之專案排程模型。其中,第一個模型為模糊多目標規劃模型,第二個模型為應用強調決策的過程,可考慮多位決策者所提供的不同專家意見的多準則暨多限制線性規劃技術(MC/sup 2/DM)所建立之模型。並進一步探討(1)如何藉由頂點法(Vertex method)自測試網路中得到模型的相關輸入資料,(2)根據緊束網路(Tight network)所輸入的資料,探討如何藉由模糊多評準決策之技巧,如以Lee's兩階段式方法(Two-stage method)的一般式,來求解上述模型。

Program evaluation and review technique (PERT) has been applied
in many areas to enhance the performance of a project. To estimate
the actual activity times over a project network, a variety of
approaches have been attempted. Among which the class of fuzzy
approaches is more suitable in realizing of real world. However,
fuzzy arithmetic operations in finding a critical path involves both
forward and backward sweeps over the same network, therefore,
multiple occurrences of activity times become inevitable. When the
extension principle method is adopted, the widened time intervals are
accrued. The widened time intervals make the searching of a critical
path difficult, and need to be eliminated.
Efficient use of resources of a project is the second issue of
this research to pursue. In fact, this problem has attracted much
attention over years, mainly focus on the deterministic approach. Due
to the uncertain nature of resource usage, a fuzzy mathematical model
is natural to be adopted to deal with the scheduling problem so that
the objectives of the shortest project completion time and resources
leveling can be unduly achieved. In this report, we formulate two
multiobjective, single project, and multiple model project scheduling
models with uncertain amount of resource usage. The first project
scheduling model is in the form of fuzzy multiobjective programming
model. The second project scheduling model is in the form of multiple
criteria and multiple constraint level linear program. We will
explore (1) how the relevant data of the model can be derived from a
test network by the vertex method. (2) how the model, along with the
input data of a tight network, can be solved by a fuzzy multiple
criteria decision making technique, such as a generalization of Lee's
two-stage method.

[科資中心編號]RA8604-2092
[題    名]
模糊神經網路專家系統之研製
Study of Fuzzy Neural Network Based Expert Systems.
[研 究 者 ]
傅心家
FU,HSIN-CHIA
[機構名稱  ]交通大學資訊工程系(NCTECSIE)
[經費來源  ]行政院國家科學委員會(NSC)
[補助金額  ]338000
[補助編號  ]NSC84-2213-E009-039
[研究開始日期]1994/08
[研究結束日期]1995/07
[頁 冊 數 ]131頁
[語    文]中文
[建檔日期  ]1997/11
[出版情形  ]完成日期:民國85年1月
[原件提供  ]行政院國家科學委員會科學技術資料中心
NSC84-2213-E009-039
[科資分類號 ]IA0407
[關 鍵 詞 ]
類神經網路;模糊邏輯;專家系統;朦朧集合;模糊控制系統;訓練演算法

Artificial neural network;Fuzzy logic;Expert system;Fuzzy set;Fuzzy
control system;Training algorithm

[摘    要]
近年來,模糊集合論已被廣泛應用在各種領域的問題上。其中,最重要的應用之一是模糊邏輯規則系統。模糊系統的優點是具有人類的推理風格,且可使用自然語言的描述方式,來處理一些不確定性的問題。然而,在一般的模糊系統處理問題的過程中,模糊規則、歸屬函數及模糊運算元的參數,常須依經驗或由嘗試錯誤設定後,方可運作。
在類神經網路方面,各式網路連結法及學習法不斷推出,其中以回傳式學習法之影響最為深遠。類神經網路的優點在具有學習能力、可平行處理、有容錯力。然而,學出來的知識卻是人類無法了解其意義,更不可能去修改。類神經網路與模糊系統的共同特色,是以數值(Numeric)的方式,來表示和處理模糊、不完整、不確定的資料。因此許多研究紛紛嘗試以二者的共同特色為基礎,擷長補短,設計具有人類思維風格、可理解、可學習的智慧型系統。大部分類神經網路的學習方式可分為兩類:回傳式學習與競爭式學習。概略地說,回傳式學習可學得較精確的知識,但卻需要耗費較長的收斂時間。競爭式學習則可作較快速的學習,然而學得的知識較不準確。因此,本計畫設計的模糊神經網路,定義Fuzzy-MIN與Fuzzy-MAX作為模糊交集與聯集之運算元。而且引用模糊推理的程序,設計出網路架構及其節點函數。學習演算則融合回傳式學習與競爭式學習的優點,以精確且快速地學習模糊邏輯控制系統所蘊涵的知識。

Recently, the theories of fuzzy sets have been applied to
various fields. One of the most important fields is the fuzzy rule
based system. Fuzzy systems provide a natural language description of
human concept for dealing with uncertain problems. However, the
decision of the fuzzy rules, the membership functions, and the
parameters of the fuzzy operations depends on human experts or trial-
and-error procedure.
In neural networks, many network topologies and learning
methodologies have been explored. Among these learning methodologies,
the backpropagation-type learning has caused enormous inference in
the research on neural networks. Though one of the characteristics of
neural network is capability of learning, the most difficult and
critical task in a neural network is how to interpret the meaning or
modify the information implied in the weights of the connections
among the neurons of the network.
The most common characteristic of neural networks and fuzzy
systems is the deal of imprecise, uncertain, and incomplete data by
numeric processing. Therefore, there are more and more research
concerning the integration of fuzzy systems and neural networks based
on their common characteristics, with the goal of combining the human
inference style and natural language use of fuzzy systems with the
learning and parallel processing abilities of neural networks. Most
of the neural networks methodologies for learning knowledge can be
divided into two categories: backpropagation-type and competitive-
type. Roughly speaking, backpropagation-type learning algorithms
learn more precisely than competitive-type algorithm does, but they
take a long time to converge. In contrast, competitive-type learning
algorithms learn much rapidly than backpropagation-type algorithm but
the knowledge learned may not be precise enough. Therefore, in this
project, the structure and the node functions of the fuzzy neural
network are designed according to the inference procedure of fuzzy
rule based system. Fuzzy-MIN and Fuzzy-MAX are defined as fuzzy
intersection and union operations, respectively. Moreover, the
learning algorithm of the fuzzy neural network is the combination of
backpropagation-type and competitive-type learning so that the
network will learn knowledge precisely and quickly.


[科資中心編號]RA8510-2124
[題    名]
模糊專家系統知識擷取環境之研製
Developing a Knowledge Acquisition System for Multiple Experts.
[研 究 者 ]
曾憲雄;黃國禎
TSENG,SHIAN-SHYONG;HWANG,GWO-JEN
[機構名稱  ]交通大學資訊科學研究所(NCTUINSG)
[經費來源  ]行政院國家科學委員會(NSC)
[補助金額  ]408000
[補助編號  ]NSC83-0408-E009-043
[研究開始日期]1994/02
[研究結束日期]1995/07
[出版日期  ]1995/07
[頁 冊 數 ]44頁
[語    文]英文
[建檔日期  ]1997/05
[出版情形  ]研究報告,44頁,民國84年7月
Report, 44p., Jul. 1995
[原件提供  ]行政院國家科學委員會科學技術資料中心
NSC84-2213-E009-066
[科資分類號 ]IA0407
[關 鍵 詞 ]
知識擷取;專家系統;朦朧集合;推論機

Knowledge acquisition;Expert system;Fuzzy set;Inference engine

[摘    要]
在本計畫中,我們研裝出一套以表格為基
底之模糊專家系統發展環境,該環境結合了知
識表格的觀念、模糊理論及個案為基底的推理
方式,它提供了友善的使用者介面、豐富的資
料形態來表達知識、智慧型的知識擷取功能以
及高效率的推理機制。為了使該環境的使用者
更精確的下決策,在推理過程及結論中有三個
因子列入計算,包括適切度、可靠性及確信度,
除此之外,為方便使用者定義適當的關係函數,
系統並提供圖型介面的曲線定義功能來協助模
糊知識的表達。

In this project, a grid-based fuzzy expert system environment,
GBFES, is proposed, which employs the concepts of repertory grid,
fuzzy theory and case-based reasoning. It offers friendly user
interfaces, abundant data types for knowledge representations,
intelligent knowledge acquisition facilities and an efficient
inference engine. To help the users in making precise decisions,
GBFES offers three factors (feasibility, reliability and certainty)
to represent the results of inference. Moreover, GBFES employs a
Bezier curve-based algorithm to help the experts in defining
membership functions.

科資中心編號: RC8501-1320 
題名: 多重屬性設備選擇之研究(I) 
The Study of Multi-Attribute Facilities Selection. 
研究者: 王泰裕 ; 吳植森 
Wang,Tai-Yue ; Wu,Chih-Sen 
機構名稱: 成功大學工業管理科學研究所(NCKUIMSG) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 625 仟元 
補助編號: NSC83-0301-H006-048 
研究日期: 1994/02 - 1995/01 
出版日期: 1995/01 
頁冊數: 84頁 
原件提供: 行政院國家科學委員會科學技術資料中心 NSC83-0301-H006-048 
科資分類號: SM0405;IB0802 
關鍵詞: 彈性製造單元 ;模糊多重屬性決策 ;模糊集合理論 ;隸屬度函數 
Flexible Manufacturing Cell (Fmc) ;Fuzzy Multi-Attribute Decision Making ;Fuzzy Set Theory ;Membership Function 
中文摘要: 在自動化已成為製造業生存的關鍵,製造廠商又得滿足顧客對產品多樣化的高度需求情況下,尋求彈性製造系統來因應此一變動,用以對應於高品質、低成本及多樣化的生產。但在全面彈性製造系統化過程中,為了防止製造系統老化與確保其擴張性,模組化之構建方式最為恰當。因此彈性製造單元快速成長,隨之對於製造單元內機器設備選購需求也迫切提升。以往的研究若不是以簡化的模式解決,便只是針對單項機器設備的選用做研究,例如機器人或夾具等的選擇方法。本研究提供一為組成製造單元所需的機器設備選擇方式,並且引進模糊集合理論,與多重屬性決策方法結合而形成之模糊多重屬性決策方法,解決採購時因對各個要素的重視程度有所不同,而引發的權數問題,至於如何選取權數隸屬函數分布,則視其能否使排列方法之解析能力與穩固性較佳而定。 
英文摘要: As automation becomes the key element for manufacturers to survive in a competitive environment and to satisfy the customers' need on product's variability and flexibility, manufactures are pursuing the flexible manufacturing system to, meet challenge in return to have the production style of high quality, low cost, and flexibility. Thus modular construction is the most appropriate method for the flexible manufacturing system to prevent the production system's aging problem during the implementation stage. Previous researchers, however, cover just single machine selection(robot or robot's gripper) or present the methodology in simplified way. This research will provide a machine selection method for constructing a flexible manufacturing cell via multi-attribute decision making method. The main objective of this methodology is to solve the problem resulted from evaluating the importance of factors inspected. Finally, the robustness and distinguishibility of membership function are used as the standard to choose the appropriate one. 
相關連結: 研究計劃摘要 
原文申請: 


[科資中心編號]RA8504-2606
[題    名]
線性模糊控制器之設計與解析
Design and Analysis of the Linear Fuzzy Controller.
[研 究 者 ]
陳誠亮
CHEN,CHENG-LIANG
[機構名稱  ]台灣大學化學工程系(NTUECHE)
[經費來源  ]行政院國家科學委員會(NSC)
[補助金額  ]408000
[補助編號  ]NSC83-0402-E002-023
[研究開始日期]1993/08
[研究結束日期]1994/07
[出版日期  ]1994/10
[頁 冊 數 ]36頁
[語    文]英文
[建檔日期  ]1997/01
[出版情形  ]研究報告,36頁,民國83年10月
Report, 36p., Oct. 1994
[原件提供  ]行政院國家科學委員會科學技術資料中心
NSC83-0402-E002-023
[科資分類號 ]IK0202;IK0204
[關 鍵 詞 ]
朦朧集合;收縮幅寬型隸屬度函數;模糊邏輯控


Fuzzy set;Shrinking-span membership function;Fuzzy logic control

[摘    要]
本報告提出一套有系統的方法來設計模糊
邏輯控制器中的隸屬度函數。此法利用「模糊
子集數」(Number of fuzzy subsets)及「收縮係數」
(Shrinking factor)兩參數來建構隸屬度函數。以此
種方法建立的隸屬度函數稱之為「收縮幅寬型
隸屬度函數」(Shrinking-span membership functions,簡
稱為SSMFs)。「收縮幅寬型隸屬度函數」的特色
是各隸屬度函數在其論域當中所占有的幅寬不
同,愈靠近設定值或平衡點的地方其幅寬愈窄
。因此,與一般研究及實際應用中常用的「等
幅寬型隸屬度函數」比較,「收縮幅寬型隸屬
度函數」更符合人類控制專家的經驗,也更合
理。
以「收縮幅寬型隸屬度函數」為基礎的模
糊邏輯控制器稱之為「收縮幅寬型模糊邏輯控
制器」,此種形式的模糊邏輯控制器在實作上
非常容易。此外,在調整「收縮幅寬型模糊邏
輯控制器」的調節變數時,工程師可以利用經
驗及控制儀器的限制作為調節的參考因素,而
不需要對這些調節變數進行繁複的最佳化運
算,所以在應用到實際的工業程序上時,不會有
理論與實際之間的鴻溝存在。故「收縮幅寬型
模糊邏輯控制器」實為一具有應用潛力的模糊
邏輯控制器。

A new means for designating membership functions in a fuzzy
logic controller is presented in this article. This method
constructs a set of membership functions systematically by using
only two parameters: number of fuzzy subsets and shrinking factors.
The membership functions generated by this method, Shrinking-Span
Membership Functions (SSMFs), have different spans in the universe
of discourse and, therefore, are more rational and more practical
from the human expert's point of view.
The fuzzy logic controller equips with such membership
functions can be implemented with little difficulties. And the
satisfactory performance of such a controller can be acquired
without laborious optimization calculation of the tuning parameters.
Since it lessens the load of computation in realizing a fuzzy
logic controller, this new approach greatly narrows the gap between
a theoretical fuzzy logic controller and a practical one, and thus
help the fuzzy logic controller more down-to-earth.

[科資中心編號]RA8504-2072
[題    名]
模糊理論應用於語音辨識系統上之研究與設計
A Study and Design of Speech Recognition System Using Fuzzy Theory.
[研 究 者 ]
胡永e
HU,YUNG-NAN
[機構名稱  ]大葉工學院電機工程研究所(DYITELEG)
[經費來源  ]行政院國家科學委員會(NSC)
[補助金額  ]220000
[補助編號  ]NSC83-0408-E212-001
[研究開始日期]1993/08
[研究結束日期]1994/07
[出版日期  ]1994/12
[頁 冊 數 ]73頁
[語    文]中文
[建檔日期  ]1997/01
[出版情形  ]研究報告,73頁,民國83年12月
Report, 73p., Dec. 1994
[原件提供  ]行政院國家科學委員會科學技術資料中心
NSC83-0408-E212-001
[科資分類號 ]IA0404;IA0406
[關 鍵 詞 ]
語音識別;朦朧集合;併圖;試探法

Speech recognition;Fuzzy set;Pattern matching;Heuristic method

[摘    要]
本研究計畫之主要目的乃是以單音辨認
(Word recognition)系統為基本架構,擬以模糊理論
(Fuzzy theory)作語音樣型比對(Pattern matching)設計,
以期達到快速準確的辨認結果。
由於人的耳朵功能類似濾波器組的效果,
因此才能判別高低音及音色,故本研究設計畫
乃以語音頻率上的特徵定義出其模糊集合(包
含模糊元素及其歸屬函數(Membership function)),而
模糊規則庫(Rule base)則以監督式訓練(Supervised
training)加上個人自我認定(Heuristic decision)來找
出最佳之推論規則。
本研究計畫係以PC/AT為主機,以C語言設計
語音讀取、訓練及模糊樣型比對等方塊功能,
來模擬辨認準確率。

The purpose of this proposal is to propose a fast and accurate
speech recognition system by using fuzzy theory. Initially, we
define the fuzzy set (including fuzzy element and membership
function) based on the characteristics of speech frequency spectrum.
Next, based on supervised training and heuristic decision to find
out the fuzzy rule and set up the data base. Finally, we use C
language to design a software to simulate the accuracy of
recognition.

科資中心編號: RA8604-2378 
題名: 分散式模糊偵測與資訊融合 
Studies of Distributed Fuzzy Detection and Information Fusion. 
研究者: 趙俊傑 
Chao,Jung-Jae 
機構名稱: 台灣海洋大學航運技術研究所(NTOUMATG) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 248 仟元 
補助編號: NSC82-0404-E019-058 
研究日期: 1993/02 - 1994/01 
頁冊數: 84頁 
原件提供: 行政院國家科學委員會科學技術資料中心 NSC82-0404-E019-058 
科資分類號: IE0509 
關鍵詞: 分散式偵測 ;信號處理 ;資訊融合 ;信號偵測 ;朦朧集合 ;Dempster-Shafer理論 
Distributed Detection ;Signal Processing ;Information Fusion ;Signal Detection ;Fuzzy Set ;Dempster-Shafer Theory 
中文摘要: Dempster-Shafer理論提供了在不確定環境下資訊融合的重要工具。此理論可證明其為Bayesian理論的一般化。Bayesian理論對一事件(Event)給予一明確的機率。Dempster-Shafer理論則在不確定的環境下給予機率的上下限。此下限機率可稱為信心量測(Belief measure)。個別偵測器的偵測信心基於此理論可融合而得新的信心。因Dempster組合法則的計算複雜度而造成實用上的問題,本研究則在一致性資訊(Consonant information)的假設下導出融合法則。在模糊分散式偵測上我們係以「信心函數」為主要工具。一方面可用之於區域偵測器上,而得含有信心程度的模糊決策;另方面又可用之於融合中心,以融合各區域偵測器。在此一計畫中我們推導出偵測與融合公式,並以兩區域偵測器為例進行信號分類的實驗。除優異的信號分辨成效外,並顯示此融合系統甚具軔性(Robustness)。 
英文摘要: Dempster-Shafer theory provides a method for information fusion where uncertain elements exist in statistical characterization of problem environment. It has been shown that this theory is essentially a generalized Bayesian theory. Instead of placing an exact probability on a given event as Bayesian theory does, upper and lower probabilities are used as likelihood bounds. The lower probability is treated as the belief measure. The degree of belief based on distinct sensor observations can be combined to form a new degree of belief which are appropriate on the basis of combined information. However, the computational complexity becomes a major problem in using Dempster's combining rule directly. In this research, we consider the consonant information only and then derive formulas for combining rules which make the fusion procedure more systematic and easier than the Dempster's approach. To examine the performance of our approach, we study the signal classification problem involving two sensors and multiple hypotheses. As a result, it shows that the proposed system greatly outperforms the one without information fusion. 


[科資中心編號]RA78-2226
[題    名]
智慧型控制系統分析與設計
Analysis and Design of an Intelligent Control System.
[研 究 者 ]
王勝德;郭德盛
WANG,SHENG-DE
[機構名稱  ]台灣大學電機工程研究所(NTUGELEG)
[經費來源  ]行政院國家科學委員會(NSC)
[補助金額  ]597000
[補助編號  ]NSC78-0408-E002-05
[研究開始日期]1988/08
[研究結束日期]1989/07
[出版日期  ]1989/09
[頁 冊 數 ]76
[語    文]中文
[建檔日期  ]1998/07
[原件提供  ]行政院國家科學委員會(NSC)
微縮片號: NSC78-0408-E002-005 1片 79幅
[科資分類號 ]IE;IE0601;IA0407
[關 鍵 詞 ]
控制系統;泛集理論;人工智慧;專家系統

Control system;Fuzzy set theory;Artificial intelligence;Expert system

[摘    要]
泛集合理論可用來處理口語化及不精確的數據,以泛集合
理論建立具有自我組織能力的專家控制系統,其優點是不需
要被控程序的數學模型,並且可用來控制複雜的程序,在程序
變動時亦具有自我調適的能力.泛集自我組織控制系統的核
心部分是規則庫及推理機構,規則庫包括控制規則與學習用
的介變規則,規則的正確性影響控制的效果,推理機構是根據
輸入找出對應的規則,由規則推導出所需的結果,推理機構在
控制及學習部分都要使用,故推理機構的精確性及運算速度
決定系統的性能.在本篇論文中,首先回顧一些泛集合的基本
理論及泛集推理的方法,然後提出一種有效的簡易的泛集推
理的演算法,其優點是節省所需的記憶體,提高運算速度,可以
應用於即時控制系統,並且加上自我組織部分時,亦可以很容
易的修改控制規則.由泛集理論發展出泛集控制器,在所需控
制的變數增加時,規則將會變得很複雜,縱然使用有效的規則
表示法,所需的記憶體也會大量增加,並且推理速度顯著降低,
針對此問題,提出多變數泛集控制系統的架構,將整個控制器
分解成數個控制單元,再將每個單元的結果以線性組合的方
式合成被控程序的控制信號.具有自我組織能力的泛集控制
器,因有自我調適能力,具有極佳的穩定性與強健性,在論文的
最後,將以實例的模擬說明泛集控制器的特性,並且驗證多變
數控制系統的架構.

This paper presents a knowledge-based intelligent control system utilizing
the fuzzy set theory. The fuzzy set theory is appropriate for dealing with the linquistic or uncertain informations. A self-organizing expert control system is developed along with a fuzzy reasoning method for a class of uncertain systems. The expert control system is consisted of a fuzzy inference engine, a rule table, a performance evaluator, and the man-machine interface. The system can also handle the multi-input muli-output case. Simulation study demonstrates the effectiveness of the proposed approach.

[科資中心編號]RA78-3135
[題    名]
通用型模糊控制器之設計及應用
Design and Application of a General-Purpose Fuzzy Controller.
[研 究 者 ]
陳誠亮
CHEN,CHENG-LIANG
[機構名稱  ]台灣大學化學工程系(NTUECHE)
[經費來源  ]行政院國家科學委員會(NSC)
[補助金額  ]205000
[補助編號  ]NSC77-0402-E002-17
[研究開始日期]1988/04
[研究結束日期]1989/03
[出版日期  ]1989/06
[頁 冊 數 ]202
[語    文]中文
[建檔日期  ]1998/07
[原件提供  ]行政院國家科學委員會(NSC)
微縮片號: NSC77-0402-E002-017 3片219幅
[科資分類號 ]IK02;IK0202
[關 鍵 詞 ]
模糊集合論;模糊控制;迴饋控制

Fuzzy set;Fuzzy controller;Feedback process control

[摘    要]
本文之主旨在於探討智慧型模糊控制器之設計與執行問題
.本文依迴饋控制各階段(伺服控制,調節控制,限制控制等)之
特性,分別設計適用之口語化控制規則及其執行方式,以達成
擬人化之控制功能.文中將設計變數(如模糊變數之隸屬度函
數,控制規則等)區分為次要與必要等類,並固定次要變數,只
保留少數主要變數(模糊變數之量化因子)供使用者依應用程
序特性調諧,以大量減少傳統模糊控制器中需設計之變數過
多的缺點.所設計模糊控制器之性能及其韌性以模擬例題測
試,均能獲得驗證.此控制器並經設計成以個人電腦為基礎之
可攜式控制器,經試用於部份模擬連續攪拌反應槽之溫度控
制,顯示出其實用性.

This thesis aims at studying the design and implementation methods for an Adaptive Intellectual Fuzzy Logic Controller (AIFLC). The AIFLC is designed to imitate the judging and control actions of the expert operator. Different linguistic control rules are proposed in various steps of feedback process control (e.g. servo control, regulatory control, contraints). In the AIFLC, most design variables are given directly to simplify the tuning procedures. Only the scaling factors of the fuzzy variables are considered performance- relevant. Some simulated examples are given to illustrate the performance and its robustness of the AIFLC. The applicability of the AIFLC is also demonstrated experimentally on a partially simulated CSTR.

[科資中心編號]RA78-2005
[題    名]
發展即時控制應用專家系統
The Development of Real-Time Expert System for Process Control.
[研 究 者 ]
王勝德
WANG,SHENG-DE
[機構名稱  ]台灣大學電機工程研究所(NTUGELEG)
[經費來源  ]行政院國家科學委員會(NSC)
[補助金額  ]146000
[補助編號  ]NSC77-0408-E002-02
[研究開始日期]1987/08
[研究結束日期]1988/07
[出版日期  ]1988/09
[頁 冊 數 ]66
[語    文]中文
[建檔日期  ]1998/07
[原件提供  ]行政院國家科學委員會(NSC)
微縮片號: NSC77-0408-E002-002 1片 65幅
[科資分類號 ]IA;IA0407
[關 鍵 詞 ]
專家控制系統;泛集合理論;專家系統

Expert control system;Fuzzy set theory;Expert system

[摘    要]
在本報告中,我們首先回顧了一些泛集合的基本理論,並探
討泛集推理方法,提出專家控制系統的架構,此架構主要是由
知識庫,推理機構和控制介面組成,知識庫是以規則為基礎的,
規則的來源可以是專家的經驗或是實際測試,推理機構則是
專家控制系統的主要部份.在本文中,我們亦提出一個適合即
時控制應用的推理方法,最後,例舉兩個範例來說明泛集專家
控制系統之設計.

In this report, we first review the fuzzy set theory and the fuzzy
reasoning methods. Then, we discussed an expert control system architecture, which is composed of a knowledge base, an inference engine and a control interface. The knowledge base is rule-based, and can be constructed from the experiences of human expert or actual experiments. The main design issue is on the inference engine. In this report, we proposed an inference method suitable for real-time control application. Two examples are also included to show the design process of an expert control system.

[科資中心編號]RA78-2060
[題    名]
以FUZZY觀念建立手寫字型識別系統之研究
Optical Character Reader for Handwritten Chinese Characters by Fuzzy Set
Theory.
[研 究 者 ]
許文星;王小川
HSU,WEN-HSING; WANG,HSIAO-CHUAN
[機構名稱  ]清華大學電機工程研究所(NTHUELEG)
[經費來源  ]行政院國家科學委員會(NSC)
[補助金額  ]2180000
[補助編號  ]NSC76-0408-E007-12
[研究開始日期]1987/04
[研究結束日期]1988/03
[出版日期  ]1988/12
[頁 冊 數 ]88
[語    文]英文
[建檔日期  ]1998/07
[原件提供  ]行政院國家科學委員會(NSC)
微縮片號: NSC76-0408-E007-012A 2片100幅
[科資分類號 ]IA;IA0406
[關 鍵 詞 ]
中文字;圖形辨認;模糊集觀念;K;指定問題;手寫字辨認

Chinese characters;Graphic recognition;Fuzzy subset theory;Entropy;Assignment
problem;Handwritten character recongnition

[摘    要]
本計畫的目的主要是研究一套新型的中文資料輸入系統,
稱為「光學中文自動識別系統」.此系統最大的優點是能夠
將手寫的中文字利用圖形識別的技術自動地輸入計算機中,
解決了現在利用鍵盤以人為方式輸入的缺點.本系統可區分
為四大部分.第一是文字影像的輸入和前處理,第二是文字筆
劃抽出,第三是文字字根切出,第四是文字的比對及識別.此四
部分經過了前二年的計畫研究已有了具體的成果.在第一年
的計畫中,利用攝影機取像,經二值化,細線化後以特徵點筆劃
抽出法得到文字筆劃,然後利用筆劃之空間分布將中文字拆
成字根,最後以筆劃之模板比對(template matching)識別中文字.第
二年之計畫中,則改用光學掃描器輸入文字影像,此外利用霍
氏轉換法抽出文字筆劃,而在字根切出部分加上背景細線化
方法,最後利用動態規劃比對法,以提高識別率.在本年度計畫
中除了保留原有的研究,對於文字筆劃的抽出部分,希望加上
模糊集(Fuzzy subset)的觀念,另外在比對方面則加上熵(entropy)的
觀念,以期得到更令人滿意的結果.
所謂模糊集是指一個集合,其元素的屬性並非絕對的(屬於
或不屬於),而是一種可能性的表示法.這種模式較符合一些常
用的觀念,如長短,高矮,異同……等等.這可能性常以一等級
來表示,而其決定的方法常是主觀的.模糊集理論的優點就在
以主觀(經驗)的表示式經由其推演程序作整體(集合本身)的
特性決策.
在本研究中,一個手寫中文字被視為一連串的筆劃所成的
集合.就形態及位置而言,由於其變動性而無法明確的將筆劃
歸類.但可藉二筆劃之相似度作為識別他們相同類型的等級.
識字問題於是轉成為:文字其對應筆劃為相同嗎?顯然這成為
一個模糊集.可用模糊集理論來解.而對於「對應筆劃」我們
則在文中證明其選取問題可歸於(Reduce to)所謂的「指定問題
」(Assignment Problem).因此,任何解「指定問題」之計算方法皆
可用來解決這選取對應筆劃的問題. 本研究所提出的方法,全部以C語言的程式規劃完成在NEC PC-98 XA個人電腦上執行.初步的結果,識別率在96.0%,是令人滿意的.

A method based on the concept of fuzzy subset for hand written Chinese
character (HCC) recognition is proposed in this project. The HCC is considered as a collective, of which the elements are the 4-directional straight line segments called strokes. Since the features (strokes) under consideration here are fuzzy in nature, the fuzzy subset concept is ultilized to the similarity measure. The recognition scheme is proven to be the same as the assignment problem. It is to average the similarities between the corresponding strokes which are chosen by solving the assignment problem to measure of the similarity between two Chinese characters. 360 HCC's in 12 character classes are used as the test patterns. The effects of the membership function on the class separability is discussed in the report. All the programs are written with C-language and is implemented on a personal computer NEC PC-98XA.

 

科資中心編號: RR88-0165 
題名: 圖形學在網路上的應用
N/A 
研究者: 賴泳伶 
Lai, Yung-Ling 
機構名稱: 景文工商專校資訊管理科(JWCTINM) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 201 仟元 
補助編號: NSC88-2115-M023-001 
研究日期: 1998/08 - 1999/07 
原件提供: 行政院國家科學委員會科學技術資料中心 NSC88-2115-M023-001 
科資分類號: CA0103 
關鍵詞: 圖論 ;圖形分割 ;網路 ;向量序列 
Graph Theory ;Graph Partition ;Network ;Vector Sequence 
中文摘要: 一個連結了n個處理機的IPC網路,可以用一個具有n個頂點的連通圖來表示。其中每個頂點代表一個處理機,而圖形中的邊擇表示處理機之間的相連性。當我們將表示IPC網路的連通圖作適度的分割,其各分割間的邊數即為此網路中所需的連結數。尋求一連通圖的最佳分割是完全的NP問題。本計畫針對此問題加上一些特別條件後,討論各種相關的問題。 
英文摘要: Let n be the number of processors and there is an IPC network connecting them. We can represent this network system by a connected graph G=(V, E) where V={P: P is a processor} and E={e: e=(P1, P2) if and only if P1 and P2 are connected}. When we partition the vertices in the graph to two sets, the edges between them are the connection needed in the corresponding IPC network. To minimize the edges between partites is NP-Complete. This research involves adding some meaningful conditions to make the problem become polynomial and discuss some related problems. 

科資中心編號: RR88-0164 
題名: Q-循環有向圖的漢米爾頓性質及連接網路相關圖形問題之研究 
The Study of Hamiltonicity on Q-Circulant Digraphs and Related Problems in Interconnection Networks 
研究者: 董立大 
Tong, Li-Da 
機構名稱: 東海大學數學系(THUSMAT) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 233 仟元 
補助編號: NSC88-2115-M029-003 
研究日期: 1998/09 - 1999/07 
原件提供: 行政院國家科學委員會科學技術資料中心 NSC88-2115-M029-003 
科資分類號: CA0103 
關鍵詞: Q循環有向圖 ;互接網路 ;Hamiltonian性質 ;圖論 
Q Circulant Digraph ;Interconnection Network ;Hamiltonian Property ;Graph Theory 
中文摘要: 在這計畫裡,我們主要針對q-循環有向圖及連續-d有向圖的漢米爾頓性質進行探討。連續-3有向圖G(3,n,q,r) 及 q-循環有向圖Gn(q,A)且n和q互質時並未完整解決。我們判斷這個問題的幾類有向圖是否具漢米爾頓性質。在另一方面,我們對連續-d有向圖及雙環網路的無向圖(The underlying graph)漢米爾頓性質進行探討且獲得一些結果,特別在雙環網路無向圖證得充要條件。 
英文摘要: In this projection, we mainly focus on the study of Hamiltonicity of q-circulant digraphs and consecutive-d digraphs. The necessary and sufficient conditions of Hamiltonicity are given for q-circulant digraphs and consecutive-d digraphs except gcd(n,q)=1 and d=3. We give interesting results for many cases. In other word, we consider the underlying graphs of consecutive-d digraphs and double loop networks. We determined Hamiltonicity for some classes of the underlying graphs of consecutive-d digraphs and gave the necessary and sufficient condition for the underlying graph of a double network to have a Hamiltonian circuit. 

科資中心編號: RR88-0157 
題名: 圖的中心、中位及其相關問題研究 
Centers, Medians and Relative Topics in Graphs 
研究者: 李海晏 
Lee, Hai-Yen 
機構名稱: 中國工商專校國際貿易科(CNICINT) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 175 仟元 
補助編號: NSC88-2119-M163-001 
研究日期: 1998/11 - 1999/07 
原件提供: 行政院國家科學委員會科學技術資料中心 NSC88-2119-M163-001 
科資分類號: CA0103 
關鍵詞: 圖論 ;中心 ;中位 
Graph Theory ;Center ;Median 
中文摘要: 考慮圖型 G=(V,E),距離 d(u,v) 表示在圖G中由點u到點v的最短路徑的長度。如果點u到其他點的最大距離最小則稱點u為圖G的中心點,圖G中所有中心點所成的集合稱為圖G的中心集,以C(G)表示。圖G中具有最小距離和的點稱為圖G的中位,所有具有最小距離和的點所成的集合稱為圖G的中位集,以M(G)表示。圖G的加權中心集Cw(G)是指加權條件下的中心集,加權中位集Mw(G)是指加權條件下的中位集。我們曾經證明強絃圖(strongly chordal graph)的加權中位集是一個團(clique)並導出求區間圖(interval graph)和塊狀圖(block graph)的加權中位集的線性演算法。在這一次的研究裡,我們修改了求區間圖和塊狀圖的加權中位集的線性演算法,獲得一個可以在線性時間求得區間圖和塊狀圖中每一點的加權距離和的演算法。 
英文摘要: Consider a graph G=(V,E). The distance d(u,v) from vertex u to vertex v in G is the length of a shortest path from u to v. A vertex u is a central vertex if the maximum distance from u to another vertex is minimum. The center C(G) of a graph G is the set of all central vertices. The median M(G) of G is the set of vertices with a minimum distance sum. The w-center Cw(G) of G is the weightd case of the center. The w-median Mw(G) of G is the weightd case of the median. We showed that the w-median of a connected strongly chordal graph is a clique. Based on the idea of the proof, we then give a polynomial time algorithm for finding the weighted median of a connected strongly chordal graph. This algorithm is then adapted to linear time algorithms for finding the weighted median problem in interval graphs and block graphs. Now, we give a linear time algorithm for finding the weighted distance sum of all vertices in interval graphs and block graphs. This algorithm is then adapted to linear time algorithms for finding the weighted median problem in interval graphs and block graphs. Now, we give a linear time algorithm for finding the weighted distance sum of all vertices in interval graphs and block graphs. 

科資中心編號: RN9101-0106 
題名: 設計並實作圖論上編碼、壓縮與解碼之新演算法(II) 
Design and Implementation of New Algorithms That Encode, Compress, and Decode Graphs (II) 
研究者: 呂學一 
Lu, Hsueh-I 
機構名稱: 中央研究院生物研究中心(國科會委託設立) 
經費來源: 行政院國家科學委員會 
補助編號: NSC89-2218-E001-014 
研究日期: 2000/08 - 2001/07 
出版日期: 2001/12 
頁冊數: 16頁 
原件提供: 行政院國家科學委員會科學技術資料中心, NSC89-2218-E001-014 
科資分類號: IA0101 
關鍵詞: 圖論 ;編碼 ;解碼 ;壓縮 ;秩序生成樹 ;輔助字串 
Graph Theory ;Encoding ;Decoding ;Compression ;Orderly Spanning Tree ;Auxiliary String 
中文摘要: 探討如何將圖形G壓縮成字串S研究大底有三個目標:(1)S的位元數越小越好(當然S要能被解壓縮還原成G)(2)編碼和解碼所需的時間越短越好(3)有關於圖形結構的查詢能夠在不解碼的情況之下快速地由壓縮碼查出來。由於這三個目標往往互相牽制,因此本計畫探討如何儘量三者兼顧的前題下對平面圖進行編碼。更仔細來說,我們研究如何在Ο(1)時間下能夠回答(a)平面圖上任意兩點是否相鄰,以及(b)平面圖上某一個點有少鄰邊。假設G有m個邊與n個節點。對於上述問題,近期內較出色的結果是由Mnuro和Raman兩位於1997年獲得,他們證明2m+8n+ο(m+n)個位元的編碼可以在Ο(m+n)的時間內取得。不久之後在1998年我們把所需的位元數縮減為2m+(5+1/k)n+ο(m+n)。本計畫經過兩年的努力終於發展出 Orderly spanning tree的新演算法工具以及一類新的Auxiliary strings,將編碼所需的位元數進一步縮減為2m+3n+ο(m+n),這個結果是目前平面圖編碼的文獻中最好的一個成果。 
英文摘要: Let G be an n-node m-edge planar graph that has no self-loops. Under the log n-bit word model of computation, we give the best known encoding S of G such that (1) S and G can be obtained from each other in O(m+n) time, (2) the degree of a node and the adjacency of two nodes in G can be determined from S in O(1) time, and (3) the neighbors of a degreed node in G can be computed from S in O(d) time. When G has (respectively, has no) multiple edges, our S has 2m+3n+o(m+n) (respectively, 2m+2n+o(n)) bits, reducing the bit counts of the encodings of Chuang, Garg, He, Kao, and Lu by at least 2n . The bit counts of our encodings are close to Tutte's information-theoretical lower bound of roughly 3:58m bits for encoding a connected plane graph without any query support. 

科資中心編號: RB9010-0171 
題名: 無AT圖距離之研究 
A Study on the Distances in Asteroidal Triple-Free Graphs 
研究者: 何錦文 
Ho, Chin-Wen 
機構名稱: 中央大學資訊工程系 
經費來源: 行政院國家科學委員會 
補助金額: 新台幣 356 仟元 
補助編號: NSC89-2213-E008-007 
研究日期: 1999/08 - 2000/07 
出版日期: 2000/09 
頁冊數: 6頁 
原件提供: 行政院國家科學委員會科學技術資料中心, NSC89-2213-E008-007 
科資分類號: IA0101 
關鍵詞: 無AT圖 ;距離 ;圖論 ;二分排列圖 ;互補比較圖 ;t擴展樹 
Asteroidal Triple Free Graph ;Distance ;Graph Theory ;Bipartite Permutation Graph ;Cocomparability Graph ;Tree T Spanner 
中文摘要: 在一圖形中,一含有三個節點的獨立集合若滿足任兩節點之間存在一條連接路徑使得該路徑不經過第三個節點與其所有鄰居,則稱此一獨立集合為asteroidal triple,並以AT簡稱之。圖形不含有AT即稱為無-AT圖。本計畫考慮無-AT圖的距離測度性質為研究主題。根據一稱為2-sweep LexBFS的演算法,我們在無-AT圖上建構一節點次序,稱該次序為「強列2-比較補圖次序」(簡稱strong 2-CCPO)。此一次序的定義乃是將原來已知的比較補圖次序加以推廣,使其滿足下列性質:圖中任意三節點若次序為u<v<w且d(u,w)≦2則d(u,v)=1或d(v,w)≦2成立。根據此一性質,我們可以獲得下列結果:(1)提供另一種新的觀點以建立無-AT圖與比較補圖之間的從屬關係。(2)證明每一個無-AT圖的真冪次圖均為比較補圖。(3)設計一新的線性演算法以識別二分排列圖。(4)證明無- AT圖具有4-擴展樹許可的特性。(5)發展較有效率的演算法以解決無-AT圖的di stance k-domination,k-stability與k- clustering等問題,其中k≧2。 
英文摘要: An asteroidal triple (AT for short) is an independent set of three vertices in a graph such that every two of them are joined by a path avoiding the closed neighborhood of the third. Graphs without asteroidal triple are called AT-free graphs. Based on a 2-sweep LexBFS algorithm, we show that every AT-free graph admits a vertex ordering, called the strong 2-cocomparability ordering (abbreviate to strong 2-CCPO), that for any three vertices u<v<w in the ordering, if d(u,w)≦2 then d(u,v)=1 or d(v,w)≦2. According to this property, we obtain the following results: (1) Provide another point of view for establishing the relationship between the classes of AT-free graphs and cocomparability graphs. (2) Prove that every proper power of an AT-free graph is a cocomparability graph. (3) Develop a simple linear time algorithm for recognizing bipartite permutation graphs. (4) Prove that every AT-free graph is tree 4-spanner admissible. (5) Design more efficient algorithms for solving the distance k-domination, k-stability, and k-clustering problems (k≧2) on AT-free graphs. 
eduction problems to be NP-hard. 

科資中心編號: RB8805-0002 
題名: 邊長問題之研究與應用
A Study of Edge Reduction Problems and Its Applications 
研究者: 杜宏毅 
Tu, Hung-Yi 
機構名稱: 靜宜大學資訊管理系(PRUMINM) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 232 仟元 
補助編號: NSC87-2213-E126-004 
研究日期: 1997/08 - 1998/07 
頁冊數: 6頁 
原件提供: 行政院國家科學委員會科學技術資料中心 NSC87-2213-E126-004 
科資分類號: IA0101 
關鍵詞: 邊長遞減 ;圖論 ;演算法 ;樹 
Edge Reduction ;Graph Theory ;Algorithm ;Tree 
中文摘要: 給定一weighted,directed,acyclic graph G. G的每一個邊都有一個值。而這個值可因所給之cost而降低。本計畫針對如何能以最少之cost而使G之最長路徑能降低最多。同時,我們也將計算出每一個邊所應降低的比率。在此計畫中我們考慮現性式以及0/1式兩種降低邊長的做法。目前我們針對DAGs, series-parallel graphs,trees已發展出有效的演算法。 
英文摘要: Let G be a weighted directed acyclic graph in which edge weights are not static quantitied, but can be reduced for a certain cost. In this project we consider the problem of determining which edges to reduce so that the length of the longest paths is minimized and the total cost associated with the reductions, linear reductions and 0/1 reductions, which model different applications. We present efficient algorithms for different classes of graph, including trees, series-parallel graphs, and directed acyclic graphs, and we show other edge reduction problems to be NP-hard. 

科資中心編號: RB8707-0034 
題名: 最佳比重2不相關集合問題之研究 
A Study on the Maximum Weight 2-Independent Set Problem 
研究者: 蕭如淵 
Hsiao, J.Y. 
機構名稱: 彰化師範大學商業教育系(NCUTBUE) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 177 仟元 
補助編號: NSC86-2213-E018-001 
研究日期: 1996/08 - 1997/07 
頁冊數: 4頁 
原件提供: 行政院國家科學委員會科學技術資料中心 NSC86-2213-E018-001 
科資分類號: IA0101 
關鍵詞: 圖形理論 ;不相關集合 ;最佳比重 ;集合問題 
Graph Theory ;Independent Set ;Maximum Weight ;Set Problem 
中文摘要: 對一俱比重(Weighed)圖形G=(V,E),S.lhkeq.V稱為G的一不相關集合(Independent set),如果任兩點a,b.in.S,a與b之間沒有邊相連。所謂的最佳比重2不相關集合問題(Maximum weight 2-independent set problem)就是要決定G上兩組無交集的不相關集合S/sub 1/與S/sub 2/使得S/sub 1/與S/sub 2/聯集後的點之比重總和要最大。最佳2不相關集合問題雖是最佳比重2不相關集合問題的一特殊情況,但在一般圖形上確是NP-complete的問題,而其應用在過去十幾年間已被廣泛而且深入的探討。在本計畫中,我們探討最佳比重2不相關集合在一些特殊圖形上俱有那些特性,並提出兩個線性時間的演算法分別在串列平行圖形(Series-parallel graph)及塊狀圖形(Black graph)上解決最佳比重兩不相關集合問題,又提出一O(.absolute.(V)/sup 2/)的演算法在區間圖形(interval graph)上解決最佳比重2不相關集合問題。 
英文摘要: An independent set of a weighted graph G=(V,E) is a subset of the weighted vertices S.lhkeq.V such that no two of the vertices in S adjacent. The maximum weighted 2-independent set (MW2IS) problem on G is to determine 2 disjoint independent sets S/sub 1/ and S/sub 2/ of G such that the summation of the vertices weight in S/sub 1/.union.S/sub 2/ is maximum. The maximum 2-independent set (M2IS) problem, which is a special case of the MW2IS problem, is NP-complete for general graphs and its applications have been extensively studied in the last decade. In this project, we studied the properties of the MW2IS on some special graphs and proposed two linear time algorithms to solve the MW2IS problem on the series-parallel graphs and the block graphs. We also proposed on O(.absolute.(V )/sup 2/) algorithm to solve the MW2IS problem on the interval graphs. 

科資中心編號: RB8705-0041 
題名: 星形聯結網路上最佳路徑找尋的組合分析 
Combinatorial Analysis of Optimal Routing in Star Networks 
研究者: 陳啟彰 
Chen, Chi-Chang 
機構名稱: 大同工學院資訊工程系(TTCTINS) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 242 仟元 
補助編號: NSC86-2213-E036-004 
研究日期: 1996/08 - 1997/07 
頁冊數: 4頁 
原件提供: 行政院國家科學委員會科學技術資料中心 NSC86-2213-E036-004 
科資分類號: IA0202 
關鍵詞: 星形網路 ;路徑搜尋 ;聯結網路 ;圖形理論 
Star Network ;Path Searching ;Interconnection Network ;Graph Theory 
中文摘要: 星形聯結網路最近引起相當多的研究,研究中證明星形聯結網路可被用來作為目前廣為使用的Hypercube網路的替代網路。本計畫和目前許多研究計畫一樣,是用來證明星形網路可以像Hypercube一樣具有許多優良特性。在本計畫中我們研究了星形網路中的平行路徑找尋問題,我們研究了點對點和一點對多點的平行路徑找尋問題。對於星形網路中一點對另外n-1點的平行路徑找尋,我們發展了一套有效率的演算法。在本計畫中我們證明出,利用此演算法可找到非常接近最短距離的n-1條平行路徑,而且我們的演算法俱有最佳的時間複雜度。我們的演算法同時可作為星形網路的垂直分割(Orthogonal partition)的一個最佳應用範例。對於點對點的平行路徑找尋問題,在本計畫中我們提供了一個有效率的演算法來求得最佳解。我們的研究結果改進了先前許多相同問題研究的結果。 
英文摘要: Star networks were proposed recently as an attractive alternative to the well-known hypercube models for interconnection networks. Extensive research has been performed showing that star networks are as versatile as hypercubes. This project is an effort in the same direction. Based on the well-known paradigms, we study the one-to-many parallel routing problem on star networks and develop an improved routing algorithm that finds n-1 node-disjoint paths between one node and a set of other n-1 nodes in the n-star network. These parallel paths are proven of minimum length within a small additive constant, and our algorithm has an optimal time complexity. This result significantly improves the previous known algorithms for the problem. Moreover, the algorithm well illustrates an application of the orthogonal partition of star networks, which was observed by the original inventors of the star networks but seems generally overlooked in the subsequent study. We also study the node-to-node parallel routing problem in the star network. We derive an efficient algorithm that constructs an optimal parallel routing in star networks. Our result improves previous result for the problem. 

科資中心編號: RA9009-0396 
題名: 圖形分割問題的數值方法 
Numerical Algorithms for Graph Partitioning Problems 
研究者: 程宣仁 
Cheng, Hsuan-Jen 
機構名稱: 清華大學數學系 
經費來源: 行政院國家科學委員會 
補助金額: 新台幣 269 仟元 
補助編號: NSC89-2115-M007-007 
研究日期: 1999/08 - 2000/07 
出版日期: 2000/08 
頁冊數: 9頁 
原件提供: 行政院國家科學委員會科學技術資料中心, NSC89-2115-M007-007 
科資分類號: CA0103 
關鍵詞: 圖論 ;分割 ;最佳切割 ;數值演算法 ;特徵值 ;下限 
Graph Theory ;Partitioning ;Optimal Partition ;Numerical Algorithm ;Eigenvalue ;Lower Bound 
中文摘要: Graph partitioning problems and NP-hard problems and very important in VLSI design. We study relations among several eigenvalue bounds and algorithms for graph partitioning problems. We design some algorithms for the problems which performs the following: first it computes the K largest eigenvalues of the affine symmetric matrix function to attain Donath-Hoffman bound; then it calculates a relaxed partition which is an array constant factor of an eigenspace associated with k eigenvalues; finally it generates an actual partition from the relaxed solution by using a method similar to Boppana's algorithm. We use a subgradient method to compute the Donath-Hoffman eigenvalue bound. Although the Donath-Hoffman bound is not tight for graph partitioning problems, our algorithm can generate optimal partitions. 

科資中心編號: RA8910-0178 
題名: Q-循環有向圖的漢米爾頓性質及連接網路相關圖形問題之研究 
The Study of Hamiltonicity on Q-Circulant Digraphs and Related Problems in Interconnection Networks 
研究者: 董立大 
Tong, Li-Da 
機構名稱: 東海大學數學系(THUSMAT) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 233 仟元 
補助編號: NSC88-2115-M029-003 
研究日期: 1998/09 - 1999/07 
原件提供: 行政院國家科學委員會科學技術資料中心 NSC88-2115-M029-003 
科資分類號: CA0103 
關鍵詞: Q循環有向圖 ;互接網路 ;Hamiltonian性質 ;圖論 
Q Circulant Digraph ;Interconnection Network ;Hamiltonian Property ;Graph Theory 
中文摘要: 在這計畫裡,我們主要針對q-循環有向圖及連續-d有向圖的漢米爾頓性質進行探討。連續-3有向圖G(3,n,q,r) 及 q-循環有向圖Gn(q,A)且n和q互質時並未完整解決。我們判斷這個問題的幾類有向圖是否具漢米爾頓性質。在另一方面,我們對連續-d有向圖及雙環網路的無向圖(The underlying graph)漢米爾頓性質進行探討且獲得一些結果,特別在雙環網路無向圖證得充要條件。 
英文摘要: In this projection, we mainly focus on the study of Hamiltonicity of q-circulant digraphs and consecutive-d digraphs. The necessary and sufficient conditions of Hamiltonicity are given for q-circulant digraphs and consecutive-d digraphs except gcd(n,q)=1 and d=3. We give interesting results for many cases. In other word, we consider the underlying graphs of consecutive-d digraphs and double loop networks. We determined Hamiltonicity for some classes of the underlying graphs of consecutive-d digraphs and gave the necessary and sufficient condition for the underlying graph of a double network to have a Hamiltonian circuit. 

科資中心編號: RA8910-0171 
題名: 圖的中心、中位及其相關問題研究 
Centers, Medians and Relative Topics in Graphs 
研究者: 李海晏 
Lee, Hai-Yen 
機構名稱: 中國工商專校國際貿易科(CNICINT) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 175 仟元 
補助編號: NSC88-2119-M163-001 
研究日期: 1998/11 - 1999/07 
原件提供: 行政院國家科學委員會科學技術資料中心 NSC88-2119-M163-001 
科資分類號: CA0103 
關鍵詞: 圖論 ;中心 ;中位 ;強絃圖 ;區間圖 ;塊狀圖 
Graph Theory ;Center ;Median ;Strongly Chordal Graph ;Interval Graph ;Block Graph 
中文摘要: 考慮圖型 G=(V,E),距離 d(u,v) 表示在圖G中由點u到點v的最短路徑的長度。如果點u到其他點的最大距離最小則稱點u為圖G的中心點,圖G中所有中心點所成的集合稱為圖G的中心集,以C(G)表示。圖G中具有最小距離和的點稱為圖G的中位,所有具有最小距離和的點所成的集合稱為圖G的中位集,以M(G)表示。圖G的加權中心集Cw(G)是指加權條件下的中心集,加權中位集Mw(G)是指加權條件下的中位集。我們曾經證明強絃圖(Strongly chordal graph)的加權中位集是一個團(clique)並導出求區間圖(Interval graph)和塊狀圖(Block graph)的加權中位集的線性演算法。在這一次的研究裡,我們修改了求區間圖和塊狀圖的加權中位集的線性演算法,獲得一個可以在線性時間求得區間圖和塊狀圖中每一點的加權距離和的演算法。 
英文摘要: Consider a graph G=(V,E). The distance d(u,v) from vertex u to vertex v in G is the length of a shortest path from u to v. A vertex u is a central vertex if the maximum distance from u to another vertex is minimum. The center C(G) of a graph G is the set of all central vertices. The median M(G) of G is the set of vertices with a minimum distance sum. The w-center Cw(G) of G is the weightd case of the center. The w-median Mw(G) of G is the weightd case of the median. We showed that the w-median of a connected strongly chordal graph is a clique. Based on the idea of the proof, we then give a polynomial time algorithm for finding the weighted median of a connected strongly chordal graph. This algorithm is then adapted to linear time algorithms for finding the weighted median problem in interval graphs and block graphs. Now, we give a linear time algorithm for finding the weighted distance sum of all vertices in interval graphs and block graphs. This algorithm is then adapted to linear time algorithms for finding the weighted median problem in interval graphs and block graphs. Now, we give a linear time algorithm for finding the weighted distance sum of all vertices in interval graphs and block graphs. 

科資中心編號: RA8510-2006 
題名: 最佳生成樹
Optimal Spanning Trees. 
研究者: 徐力行 
Hsu,Lih-Hsing 
機構名稱: 交通大學資訊科學研究所(NCTUINSG) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 378 仟元 
補助編號: NSC84-2213-E009-009 
研究日期: 1994/08 - 1995/07 
出版日期: 1995/07 
頁冊數: 46頁 
原件提供: 行政院國家科學委員會科學技術資料中心 NSC84-2213-E009-009 
科資分類號: IA0101;IA0402 
關鍵詞: 最佳生成樹 ;圖論 ;無向網路 ;最佳化 
Optimal Spanning Tree ;Graph Theory ;Undirected Network ;Optimization 
中文摘要: 本計畫主要研究:在一無向網路上,針對各種不同的要求找尋對應之最佳生成樹(Optimal spanning trees)。此問題無論在理論上或實際應用上都極具價值。例如:給定一個無向圖G=(V,E),G上的每一條邊e對應w(e)代表e的造價,G上的任何一生成樹T的成本(Cost)定義為cost(T)=.SUMMATION. e.in.T(w(e))。找尋一個生成樹T*使其成本比G中任一其他生成樹成本都要小的問題,其重要性一直為世人所公認。若G上的每一條邊e附有c(e)與p(e)分別代表其造價(成本)及可靠度(利益),G上任一生成樹T的權數(Weight)定義為.SUMMATION. e.in.T(c(e))/.PI. e.in.T(p(e))。找尋G中權數最小的生成樹,稱為最佳本益比樹(Optimal cost-reliability ratio spanning tree),其意義在於找尋低成本而可靠度高的生成樹,在網路設計及其他實際應用中十分重要。本報告分為三個部分:第一個部分為最佳本益比之生成樹與相關問題。第二個部分是雙參數標的之最佳生成樹的元素擾亂問題。第三個部分為依據SKT可靠度在BSP有向圖形上尋找最重要邊。 
英文摘要: The primary purpose of this project is to find optimal spanning trees with different objects. This problem is one of the most important subject both in theoretical or practical world. For example, given a undirected graph G=(V,E), there exists a weight w(e) which corresponds to each edge e of G. The cost of any spanning tree T of G, cost(G), is defined to be cost(T)=. SUMMATION. e.in.T(w(e)). People have recognized the importance of finding a spanning tree T* such that cost of T*, cost(T*), is less than those of other spanning trees. Suppose that two weights c(e) and p(e) are associated with any edge of an undirected graph G=(V,E) where c(e) reflects the cost and p(e) the reliability of edge e, respectively. The weight of any spanning tree T is defined to be .SUMMATION. e.in.T(c(e))/.PI. e.in.T(p(e)). The least weighted spanning tree of G is called the optimal cost-reliability ratio spanning tree. Finding such a tree is to find a spanning tree with the least cost and the highest reliability. It is important in designing networks and other applications. There are three parts in this report. (1) On minimal cost- reliability ratio spanning trees and related problems. (2) Element perturbation problems of optimum spanning trees with two-parameter objectives. (3) Finding the most vital edge with respect to SKT reliability in BSP digraphs. 

科資中心編號: RA8507-2087 
題名: 固定方向之幾何問題研究 
Shortest Paths among Polygonal Obstacles in Fixed Orientations. 
研究者: 鄭進和 
Cheng,Chin-Ho 
機構名稱: 輔仁大學資訊工程系(FJUSINE) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 98 仟元 
補助編號: NSC82-0113-E030-007-T 
研究日期: 1993/02 - 1994/01 
出版日期: 1994/03 
頁冊數: 22頁 
原件提供: 行政院國家科學委員會科學技術資料中心 NSC82-0113-E030-007-T 
科資分類號: IA0402 
關鍵詞: 演算法 ;最短路徑問題 ;固定方向 ;圖論 
Algorithm ;Shortest Route Problem ;Fixed Orientation ;Graph Theory 
中文摘要: 本計畫在研究平面上固定方向之最短路徑問題,研究中有一假設,即事先給定一組有限個固定方向,然後在平面上擺放一些障礙物(以多邊形表示之)與兩點---一是起點,另一是終點(起點與終點皆不在障礙物內部),尋找一條連接起點與終點,繞過障礙物的路徑,而且此路徑上之每一線段其方向為給定的固定方向之一,從所有可行的路徑當中,挑出一條最短的路徑。在研究過程中,首先找出一些關於固定方向路徑的幾何性質,再利用這些性質與圖形理論最短路徑的方法,而得到一尋求固定方向最短路徑的演算法,這演算法花O(nlog/sup 2/n)時間與O(nlog n)記憶體,其中n是障礙物周界頂點的數目。 
英文摘要: The problem of finding a shortest path among simple polygonal obstacles in fixed orientations is considered in this study. This problem may be stated as follows: Given a fixed set of orientations and a set of obstacles in the plane, find a shortest path from a point s to a point t which avoids all obstacles and is composed of line segments with only the given orientations. The path may touch an obstacle but does not cross an obstacle. By using a graph- theoretical approach, we obtain an O(nlog/sup 2/n) algorithm for finding such a path, where n is the number of the vertices of the obstacles. 

科資中心編號: RA8507-2005 
題名: 聯合圖上演算法之探討 
A Study of Algorithms on Cographs. 
研究者: 余明興 
Yu,Ming-Shing 
機構名稱: 中興大學應用數學研究所(NCHUAPMG) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 294 仟元 
補助編號: NSC83-0408-E005-003 
研究日期: 1993/08 - 1994/07 
出版日期: 1994/07 
頁冊數: 56頁 
原件提供: 行政院國家科學委員會科學技術資料中心 NSC83-0408-E005-003 
科資分類號: IA0101;IA0402 
關鍵詞: 圖論 ;聯合圖 ;演算法 
Graph Theory ;Cograph ;Algorithm 
中文摘要: 在應用數學和計算機科學的許多不同領域中,往往會自然地產生出一類圖形---聯合圖形(Cographs)。在本計畫中,我們專注於加權的情況並展示一個時間複雜度為O(n)的循序演算法和一個在EREW PRAM上需要時間O(log n)和處理器O(n/log n)的平行演算法來解加權聯合圖形(Weighted cographs)上之最重獨立集(Maximum weight independent set)的問題。而我們可以使用類似的演算法在相同的時間複雜度及處理器之限制來解決加權聯合圖形上的最輕點覆蓋(Minimum weight vertex cover)、最重完全子圖(Maximum weight clique)和最輕獨立支配集(Minimum weight independent dominating set)等問題。同時我們證明出在加權聯合圖形上的一個最輕極大精簡集(Minimum weight maximal irredundant set)就是一個最輕支配集(Minimum weight dominating set),反之亦然。並且我們提供一個在EREW PRAM上需要時間O(log n)和處理器O(n/log n)的平行演算法來解這兩種問題。 
英文摘要: The class of cographs, or complement-reducible graphs, arises naturally in many different areas of applied mathematics and computer science. In this project we concentrate on the weighted case and present an O(n) time sequential algorithm and a parallel algorithm of O(log n) time and O(n/log n) processors on the EREW PRAM model to solve the maximum weight independent set problem on weighted cographs. Using such algorithms we can easily solve the minimum weight vertex cover problem, the maximum weight clique problem and the minimum weight independent dominating set problem on weighted cographs with the same bounds of time and processors. And we show that minimum weight maximal irredundant set is also a minimum weight dominating set on a weighted cograph and vice versa. Also we provide a parallel algorithm of O(log n) time and O(n/log n) processors which can solve these two problems simultaneously on weighted cographs. 

科資中心編號: RA8407-1046 
題名: 圖論上的平行演算法 
Parallel Algorithms on Graph Theory. 
研究者: 林朝枝 
Lin,Chau-Jy 
機構名稱: 交通大學應用數學系(NCTSAPM) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 326 仟元 
補助編號: NSC83-0208-M009-027 
研究日期: 1993/08 - 1994/07 
出版日期: 1994/08 
頁冊數: 18頁 
科資分類號: CA0103 
關鍵詞: 演算法 ;圖論 ;平行計算機 ;圖形 ;郝米頓週期圈 ;韻律演算法 ;線性陣列 ;處理單元 
Algorithm ;Graph Theory ;Parallel Computer ;Graph ;Hamiltonian Cycle ;Systolic Algorithm ;Linear Array ;Processing Element 
中文摘要: 給予任意一個圖G=<V,E>,V共有n個點,我們提出一個韻律演算法來解G上是否存在郝米頓循環週期圈問題。所使用的計算模式為一個含有(n-1)個處理單元的環狀線性陣列。每兩個相鄰的處理單元都有三條資料傳輸線連接著。當存在某個循環週期圈時,所需的時間步驟顯得相當合理。可是在循環週期圈不存在時,將要使用較多的時間步驟來證明其不存在的事實性。在此模式中,每個處理單元都執行相同的簡單結構,而且每個處理單元都執行相同的指令。我們希望此韻律演算法可以應用來求旅行推銷者的問題之解。 
英文摘要: For an arbitrary graph G=<V,E> with .absolute.(V)=n, we present a systolic algorithm to solve the Hamiltonian cycle problem on G. The computational model used is a linear array consisting of(n-1) identical processing elements (PEs). There are three communication links between any two neighboring PEs. When a cycle exists, it is obtained within a reasonable time steps. However, it requires more time steps to confirm the absence of a cycle. Each PE has identical and simple structure. Any PE executes same instructions at a time step. We hope that this parallel algorithm can be applied to obtain the solution of the travelling salesman problem. 

科資中心編號: RA8404-2028 
題名: 圖形區域性之研究 
A Study on Graph Localities. 
研究者: 張瑞雄 
Chang,Ruay-Shiung 
機構名稱: 台灣工業技術學院資訊管理技術系 (NTITINM) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 311 仟元 
補助編號: NSC83-0408-E011-003 
研究日期: 1993/08 - 1994/07 
出版日期: 1994/08 
頁冊數: 27 
科資分類號: IA;IA0403 
關鍵詞: 圖論 ;圖形區域性 ;深度搜尋 ;廣度搜尋 ;分割而 征服 ;分枝而限制 ;動態規劃 
Graph Theory ;Graph Locality ;Depth First Search ;Breadth First Search ;Divide-And-Conquer ;Branch-And-Bound ;Dynamic Programming 
中文摘要: 由於電腦軟體與硬體的快速發展使得多處 理系統、電腦網路及分散式系統愈來愈受到歡 迎而且更負擔得起。在這些系統中,合適的演 算法仍是這些系統能否有效的使用的重要因素 之一。本研究以一個圖形的寬搜尋樹為基礎, 來探討圖形的區域性,根據圖形的區域性我們 提出了可調式的平行及主從模式的演算法來解 決平面圖的測試問題、最小值擴展樹問題和最 大的完全子圖問題。所謂可調式的意思是不管 系統有多少的處理器可同時用,此類演算法仍 可用。 
英文摘要: The advances in hardware and software technologies have made multiprocessor systems, computer networks, and distributed systems more and more popular and affordable. In these systems, suitable algorithms are still one of the crucial factors for such systems to be useful. In this research, we study the graph localities based on its breadth first search trees. We propose scalable parallel and client-server algorithms to solve the planarity testing problems, the minimum spanning tree problem, and the maximum clique problem. The scalability means that no matter what the available number of processors is, the algorithms can still work. 

科資中心編號: RA8404-2008 
題名: 平行計算中的樹收縮技術及其應用之探討 
A Study of the Tree Contraction Technique and Its Applications in Parallel Computation. 
研究者: 余明興 
Yu,Ming-Shing 
機構名稱: 中興大學應用數學系(NCHSAPM) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 140 仟元 
補助編號: NSC82-0408-E005-066 
研究日期: 1993/02 - 1994/01 
出版日期: 1994/01 
頁冊數: 55 
科資分類號: IA;IA0101 
關鍵詞: 平行計算 ;樹收縮 ;圖論 
Parallel Computing ;Tree Contraction ;Graph Theory 
中文摘要: 在平行計算(Parallel computation)的領域中,樹 收縮(Tree contraction)技術可以用來解決許多問題 。例如,我們可以用它來解決某些運算式求值 (Expression evaluation)及語言識別(Language recognition) 的問題。另外,在圖論(Graph theory)中的若干問題 也可以用樹收縮的技術來解決。如在樹圖(Tree) 、串聯並聯圖(Series parallel graph)、平面圖 (Planar graph)等,上面有許多問題都可用樹收縮的 技術來解決。傳統的樹收縮都是針對二元樹(Binary tree) 在作運算。在本計畫中,我們發展出一種樹收 縮技術可以直接在一般樹上作運算,這種方式 可以讓我們在原來的樹上思考問題,而不必先 轉換到二元樹。運用這種樹收縮技術,我們解 了以下四個加權樹(Weighted tree)上面的問題:(1) 最小加權點涵集(Minimum weight vertex cover),(2)最小 加權支配集(Minimum weight dominating set),(3)最大加 權獨立集(Maximum weight independent set),(4 )最大加權配對(Maximum weight matching)。這些平行演算法的 複雜度都是 O(logn)時間及O(n/logn)的處理器,計算 模型為EREW PRAM。 
英文摘要: In the realm of parallel computation, the technique of tree contraction can be applied to solve many problems. For instance, we can use it for some problems about expression evaluation and language recognition. Moreover, some problems in graph theory can also be solved by using tree contraction. For example, many problems on trees, series parallel graphs, and planar graphs can be solved by using tree contraction. Traditionally, tree contraction works only on binary trees. In this project, we have developed a technique of tree contraction which can be applied directly on general trees. Such a technique will allow us to deal with problems on the original tree, instead of considering the problem on a binary tree. By using this tree contraction technique, we solved the following problems on weighted trees: (1) minimum weight vertex cover, (2) minimum weight dominating set, (3) maximum weight independent set, and (4) maximum weight matching. All of the above parallel algorithms are of O(logn) time and O(n/logn) processors under the EREW PRAM model. 

科資中心編號: RA8301-2006 
題名: 單一步圖形搜尋問題的研究 
A Study on the Single Step Graph Searching Problems. 
研究者: 李家同 ; 唐傳義 
Lee,Char-Tung ; Tang,Chuan-Yi 
機構名稱: 清華大學資訊科學研究所(NTHUCOS) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 383 仟元 
補助編號: NSC82-0408-E007-031 
研究日期: 1992/08 - 1993/07 
出版日期: 1993/07 
頁冊數: 48 
科資分類號: IA;IA0403 
關鍵詞: 圖形搜尋 ;單一步圖形搜尋 ;個數最多不相關二組集合 ;圖論 ;區塊圖 ;瓶頸成本 ;混合成本 
Graph Searching ;Single Step Graph Searching ;Maximum Two-Independent Set ;Graph Theory ;Block Graph ;Bottleneck Cost ;Hybrid Cost 
中文摘要: 在這篇報告中,我們在俱比重的區塊圖形 (Weighted block graphs)上探討了三種不同類型的單 一步圖形搜尋問題(Single step graph searching problem)。它們分別是和最小化型(簡稱S-型)、瓶 頸值最小型(簡稱B-型)及混合型(簡稱H-型)等三 種不同衡量指標下的俱比重單一步圖形搜尋問 題。我們證明了這三大類的單一步圖形搜尋問 題在一般的圖形上均是NP-困難的問題。但是若 考慮區塊圖形時,這三大問題則有多項式的演算法。我們的主要技巧乃將一個區塊圖先轉換 成所謂的切點樹形結構圖(Cut-vertex tree structure) 上來求解。我們首先證明了S-型單一步圖形搜 尋問題與俱比重的最大2-獨立集問題可以在線 性的時間(Linear time)作互相的轉換。然後我們 藉由在區塊圖形上解最佳俱比重2-獨立集問題 的答案來解決區塊圖形上S-型的單一步圖形搜 尋問題,此演算法所花費的時間為線性的。而 有關B-型問題,我們的作法是:第一階段先計算所需的可能瓶頸成本(Bottleneck cost),接下來於第 二階段利用階段一所得之瓶頸限制下,再利用S- 型的演算法去求出搜尋計畫(Searching plan)。最 後綜合S-型及B-型兩問題的結果,我們利用了二 元搜尋(Binary search)技巧提出了一個時間為 O(.absolute.(E)log. absolute.(V))的演算法去解決區塊 圖形上的H-型單一步圖形搜尋問題,其中V和E分 別表點(Vertices)和邊(Edges)的集合。 
英文摘要: In this report, three types of the problems for the single step searching on weighted graphs are investigated. They are the summation minimization (S-type, in short), the bottleneck minimization (B- type, in short), and the hybrid (H-type, in short) weighted single step searching problems. The main technique of our approach is that we transform the input block graph to the corresponding cut- vertex tree structure, and then work on the cut-vertex tree structure in the rest phases. First, all of the three problems are shown to be NP-hard on general graphs but polynomial solvable on block graphs. The S-type problem is proved to be linearly equivalent to the optimum weight 2-independent set problem. Then we solve the S-type problem on a block graph G in linear time by solving the optimum weight 2-independent set problem on G. To solve the B-type problem, the first phase computes the bottleneck cost and the second phase constructs the searching plan by applying the S-type algorithm using the bottleneck cost derived in the first phase. Finally, an O(.absolute.(E)log.absolute.(V)) time algorithm using the binary search strategy for solving the H-type problem on weighted block graphs is proposed, where V and E are the sets of vertices and edges, respectively. 

科資中心編號: RA8301-1016 
題名: 圖形端點序列的跳躍數最佳化之研究 
A Study on Jump Number Optimization for Graph Vertex Sequence. 
研究者: 張瑞雄 
Chang,Ruay-Shiung 
機構名稱: 台灣工業技術學院資訊管理技術系 (NTITINM) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 318 仟元 
補助編號: NSC82-0208-M011-002 
研究日期: 1992/08 - 1993/07 
出版日期: 1993/08 
頁冊數: 36 
科資分類號: CA;CA0103 
關鍵詞: 演算法 ;圖學理論 ;部份次序理論 ;NP-完全 
Algorithm ;Graph Theory ;Partial Order ;Np-Completeness 
中文摘要: 在本篇論文,我們希望找到圖形端點的線 性排列,使得在此排列裡面,左右鄰居在原圖中 能夠相鄰(Adjacent)的愈多愈好(跳躍數愈少愈好) ;或者左右鄰居在原圖中能夠不相鄰(Nonadjacent) 的愈多愈好(跳躍數愈多愈好)。此二問題都是NP-完全問題。我們研究它們 在一些特殊圖形上的複雜度,並提出一個簡單 的樹狀圖形的最小跳躍數的演算法和Proper interval graph上的最大跳躍數的演算法。 
英文摘要: In this paper, we study the problems of finding a linear ordering of vertices sequence of a graph such that the number of consecutive two vertices that are adjacent is maximized (jump minimization problem) or the nonadjacency between consecutive vertices is maximized (jump maximization problem). Both problems are shown to be NP-hard and are investigated with respect to some special graphs. In particular, the jump minimization problem is resolved for trees with a simpler algorithm and the jump maximization problem is solved for bipartite graphs, proper interval graphs, and series-parallel graphs. 
algorithm using the binary search strategy for solving the H-type problem on weighted block graphs is proposed, where V and E are the sets of vertices and edges, respectively. 

科資中心編號: RA81-2013 
題名: 某些特殊圖形上的配對問題之探討 
A Study of the Matching Problem on Some Special Graphs. 
研究者: 余明興 
Yu,Ming-Shing 
機構名稱: 中興大學應用數學研究所(NCHYAPMG) 
經費來源: 行政院國家科學委員會(NSC) 
補助金額: 新台幣 298 仟元 
補助編號: NSC81-0415-E005-03 
研究日期: 1991/08 - 1992/07 
出版日期: 1992/07 
頁冊數: 26 
科資分類號: IA;IA0101;IA0403 
關鍵詞: 配對問題 ;圖形理論 ;演算法 ;平行演算法 
Matching Problem ;Graph Theory ;Algorithm ;Parallel Algorithm 
中文摘要: 在圖形理論中,配對問題(Matching problem)一 直是很多人有興趣的問題,也有許多重要的應 用。因為研究一般圖形(General graph)上的配對問 題已很久沒有大的進展,所以近來有很多研究 是去解決各種特殊圖形(Special graph)上的配對問 題。本計畫首先介紹配對問題及此問題在各種 特殊圖形上的演算法,我們將分成最大配對問 題(Maximum matching problem)和最大加權配對問題 (Maximum-weight matching problem)兩方面來介紹。然後 我們發展一個線性時間的演算法,去解決聯合 圖形(Cograph)上的最大配對問題。 
英文摘要: In the graph theory, matching problems have been interesting to many people for a long time, and it has many important applications. since there have been no significant improvements for this problem on general graphs for a long time, many studies try to solve this problem on various special graphs recently. In this project, we first describe the matching problem and algorithms on special graphs, we shall describe both the maximum matching and the maximum-weight matching problems. Then we develop a linear time algorithm for the maximum matching problem on cographs. 

[科資中心編號]RA8401-2011
[題    名]
多異質資料庫系統整合之研究
Research on Integration of Heterogeneous Databases.
[研 究 者 ]
李強
LEE,CHIANG
[機構名稱  ]成功大學資訊工程研究所(NCKUIFEG)
[經費來源  ]行政院國家科學委員會(NSC)
[補助金額  ]521000
[補助編號  ]NSC82-0408-E006-418
[研究開始日期]1993/02
[研究結束日期]1994/01
[出版日期  ]1994/03
[頁 冊 數 ]89
[語    文]中文
[建檔日期  ]1995/10
[科資分類號 ]IA;IA0304
[關 鍵 詞 ]
異質資料庫;結構整合;資料差異;定義域不匹配

Heterogeneous database;Schema integration;Data conflict;Domain mismatch

[摘    要]
隨著資源共享的需求日益增加,各個獨立 資料庫系統間
的合作變成了一個重要的課題。 該問題已成了一門學問,稱
之為多資料庫系統, 或者是邦聯式資料庫系統。目前,已有許
多人 投入這領域(ACM90、COM91、IMS91、REC91)。在本計 畫裡,我們對一些常見且重要的問題加以分類 探討。同時,我們提出
了一種多資料庫系統架 構。根據這個架構,我們深入探討了
三個問題, 他們分別是值域不相容、Entity identification和 Schema
conflict。並提出我們的解法。對於值域不相容問題,我們延伸了過去的 看法,提出一種新觀念,稱之為Mismatch heirarchy。 利用這個觀念,我們分別針對Selection Join和Update 運算提出新的演算法。這些演算法的目的是為 了讓這些運算在有值域不相容問題存在時也能 夠順利執行。至於Entity identification的問題,我們主張以 比對Common attribute來計算Entity的相似程度。為 了增進此法的精確度,我們提出一種Common attribute的分類方式。同時,我們也談到我們
的 計算方式與新的Join定義。最後,我們針對Schema conflict中一個資料庫 的Data,在另一個資料庫裡被以Metadata來表示的 問題提出解法。在此,我們以SQL為例,來說明我 們的方法。

Interoperability among existing autonomous database is an issue of
increasing importance due to the demand of sharing resources among computer systems. A system having that capability is called a multidatabases system or federated databases system. Various difficult problems have been identified in this category [ ACM90, COM91, IMS91, REC91 ] . Some of the most often seen issues are classified in this thesis. To resolve those problems, we propose solutions for three important and often seen issues. They are domain mismatch problems, entity identification problems, and schematic conflict problems.
For domain mismatch problems, we propose solutions to resolve relational
operations, such as selection, update, and join operation, based on a new
concept called mismatch hierarchy.
For entity identification problems, we calculate the similarity degree
of entities by comparing their common attributes. To improve result accuracy, we classify them into the following three categories, multi-lateral attribute, DB-invariant attribute, and domain-mismatched attribute. Also, we present a quantitative model to resolve the entity identification problems. Based on this model, we calculate the similarity degree of entities and extend the traditional join. A new join operation, called .delta.-instance-join, is also presented. As for schematic conflict problems, we focus on the problem that data in one database correspond to metadata (schema elements) of another database. We propose a method which utilizes query transformation techniques to this
problem.

[科資中心編號]RA8404-2033
[題    名]
關聯式及物件式資料庫的整合
Integrating Relational and Object Databases.
[研 究 者 ]
陳良弼
CHEN,ARBEE L.P.
[機構名稱  ]清華大學資訊科學研究所(NTHUCOS)
[經費來源  ]行政院國家科學委員會(NSC)
[補助金額  ]462000
[補助編號  ]NSC83-0408-E007-030
[研究開始日期]1993/08
[研究結束日期]1994/07
[出版日期  ]1994/09
[頁 冊 數 ]42
[語    文]英文
[建檔日期  ]1996/01
[科資分類號 ]IA;IA0304
[關 鍵 詞 ]
異質資料庫;關聯式資料庫;物件式資料庫;查詢 轉換;整體查
詢處理

Heterogeneous database;Relational database;Object database;Query
transformation;Global query processing

[摘    要]
本次計畫中我們探討了關聯式及物件資料 庫間整合的
問題及技術,在關聯式資料庫上,物 件式界面的建構技術、資
料庫間綱目的整合及 整體查詢處理方面,獲得相當的成果;我
們將這 些成果分別說明在以下三部分:
第一部分:我們提出在關聯式資料庫上物 件界面的構建
技術,利用我們所發展的物件觀 定義語言以及推導規則,完成
物件觀的定義與 設計,使得物件可以透過物件查詢語言來取
得, 並且,我們也提出了查詢語言的轉換技術,使得 使用者根
據這個物件界面所提出的查詢,能確 實的轉換成相對應的關
聯式查詢,進而擷取到 正確的資料,並加以適當的整合。我們
並把這 個界面實作於SYBASE關聯式系統上。
第二部分:這部分提供了一個物件綱目整 合的完整方案
。由不同資料綱目整合成的整體 綱目可以提供一致的存取
界面及高層次的資料 通透性。基於資料庫管理者所指定的
對應聲明 (Corresponding assertions)我們提供一組整合運算子 來
重整成員綱目及解決成員綱目間衝突的問 題,以達成整合的
目的。其中,重點在於整合後 的綱目仍然保持成員綱目原有
的資訊,更由於 不同的資料庫間的整合,產生了更有價值的整
合資訊。
第三部分:在此部分我們提出基於整體綱 目整合下的整
體查詢處理。利用整體綱目與成 員綱目間的對應資訊,整體
查詢可被分解成一 組相對應到關聯成員綱目的子查詢。我
們提出 了整體查詢的步驟方法與結果整合技術,並定 義一套
流程控制語言(Flow control language),來處理 分解後子查詢的執
行及其間執行順序之協調控 制與查詢結果之整合。此外,我
們在子查詢的 分解處理,我們根據整合資訊的特性提供了最
簡化的處理,來簡化不必要的子查詢及查詢條 件。

In this project, we researched into problems in constructing a
heterogeneous database system from relational and object databases. We have conquered the issues of constructing object front-end on top of relational databases, schema integration among component databases and global query processing. We present our research results in the following three parts.
Part I: In the first part, we propose the technique of constructing an
object interface on top of relational databases. By using Object View
Definition Language (OVDL) and object view derivation rules, users can derive the relational schema into a corresponding object view and then can retrieve data by object query languages. Moreover, transformation strategies from object queries to corresponding relational queries are also presented. This object front-end has been implemented in SYBASE relational database system. Part II: In this part, strategies of integration of heterogeneous object schemas are proposed. A global schema created by integrating schemas of the component databases can provide a uniform interface and high level location transparency for the users to retrieve data. Based on the corresponding assertions specified by database administrators, a set of integration operators can be used to reconstruct the component schemas for resolving the conflicts and do the integration. The principle of our integration strategy is to keep the data of component databases retrievable from the global schema without losing information. Moreover, more informative query answers may be derived from the heterogeneous databases due to schema integration. Part III: In this part, global query processing technique based on our schema integration is developed. By using the provided mapping information between global schema and component schemas, the global query is decomposed into a set of subqueries which can be executed in the corresponding local databases. The steps of global query decomposition are proposed as well as the integration of partial results. A Flow Control Language (FCL) is then defined to specify the execution flow of the subqueries and result integration. According to the mapping information of schema integration, query optimization techniques are considered in the specification of the execution flow.

[科資中心編號]RA8504-2025
[題    名]
異質性分散式資料庫中資料庫概要和異動轉換
之研究
Schema Transformation and Transaction Translation in HDDB
Environments.
[研 究 者 ]
于厚澤
YU,HOU-TZER
[機構名稱  ]國防管理學院資訊管理系(NDMGIFM)
[經費來源  ]行政院國家科學委員會(NSC)
[補助金額  ]183000
[補助編號  ]NSC83-0415-E123-001
[研究開始日期]1993/08
[研究結束日期]1994/07
[出版日期  ]1994/06
[頁 冊 數 ]116頁
[語    文]英文
[建檔日期  ]1997/01
[出版情形  ]研究報告,116頁,民國83年6月
Report, 116p., Jun. 1994
[原件提供  ]行政院國家科學委員會科學技術資料中心
NSC83-0415-E123-001
[科資分類號 ]IA0304
[關 鍵 詞 ]
異質資料庫;詢問透明性;查詢反應時間

Heterogeneous database;Inquiry transparency;Query response time

[摘    要]
在本研究中,利用異質性資料庫五層綱目
整合架構來建置一個異質性資料查詢與綱目整
合的應用程式介面,藉由檔案轉換與傳輸的方
式,建立鬆耦合(Loosely coupled)與緊耦合(Tightly
coupled)異質性資料庫系統架構,使能達到異質性
資料透通查詢的目標。
再利用不同系統架構下的限制條件,來分
析鬆耦合與緊耦合資料庫的擷取反應時間;最
後,利用遠端拷貝(Remote copy)指令的方式,來表現
異質性與同質性資料庫系統間擷取資料之時間
效益分析。
本報告共分四大部分,第一部分異質性分
散式資料庫系統概要探討,主要說明本研究之
緒論、現有系統架構描述及異質性資料庫的主
要研究方向與需求;第二部分鬆耦合與緊耦合
資料庫概要介紹,解釋五層綱目整合架構及鬆
、緊耦合資料庫的概念,以做為實作驗證的理
論基礎;第三部分系統應用與效益比較,則利用
軟體介面方法,實際設計程式並分析本研究在
鬆、緊耦合資料庫兩類不同架構下的查詢時間
差異;並將相同的方法應用在同質性資料庫的
架構中,進行查詢效益分析,觀察其中之差異性;
第四部分結論,總結成果並針對未來研究方向
給予適切之建議。

In this research, we construct an application program
interface that consisting of heterogeneous data queries and
integrated schemas in the architecture of five-level heterogeneous
database system.
The interface was implemented for both loosely coupled and
tightly coupled heterogeneous testbed database system to achieve the
objective of heterogeneous data inquiry transparency by file
transformation and transmission.
The performance evaluation had been made in different system
architectures by analyzing the response time of accessing Loosely
coupled and Tightly coupled database with various constrained
conditions.
Finally, we use remote copy command to perform time efficiency
analysis of accessing data in heterogeneous and homogeneous database
systems.

[科資中心編號]RA8504-2042
[題    名]
多資料庫系統的查詢處理與最佳化
Query Processing and Optimization in Heterogenous Database System.
[研 究 者 ]
陳良弼
CHEN,ARBEE L.P.
[機構名稱  ]清華大學資訊科學研究所(NTHUCOS)
[經費來源  ]行政院國家科學委員會(NSC)
[補助金額  ]369000
[補助編號  ]NSC83-0408-E007-029
[研究開始日期]1993/08
[研究結束日期]1994/07
[出版日期  ]1994/09
[頁 冊 數 ]70頁
[語    文]英文
[建檔日期  ]1997/01
[出版情形  ]研究報告,70頁,民國83年9月
Report, 70p., Sep. 1994
[原件提供  ]行政院國家科學委員會科學技術資料中心
NSC83-0408-E007-029
[科資分類號 ]IA0304
[關 鍵 詞 ]
異質資料庫;查詢最佳化;交易處理;分散式查詢

Heterogeneous database;Query optimization;Transaction processing;
Distributed query

[摘    要]
在本計畫中,我們對於利用不確定資料來
解決資料不一致性以及查詢處理與最佳化的問
題已得到一些結果。茲將這些研究成果分成五
個部分。其中,第一部分探討在異質資料庫中
查詢最佳化之區域性研究。第二部分提出在處
理分散式查詢時利用One-shot半連結之處理策略
。第三部分考慮如何估計機率部分值
(Probabilistic partial values)中的機率值。第四部分
提出一正確且完全的擴充關聯式代數來處理不
確定資料。第五部分探討一具高並行度的長持
續型交易處理的方法。

In this project, we have completed some works in resolving the
data inconsistency by using uncertain data, the query processing and
optimization problems in distributed multidatabase environments.
These studies are divided into the following five parts. The first
part devotes to a localized approach to query optimization in
heterogeneous database systems. In part two, we propose one shot
semi-join execution strategies for processing distributed queries.
The third part considers the problem in determining probabilities
for probabilistic partial values. In part four, we present a sound
and complete extended relational algebra for exclusive disjunctive
data. The last part investigates the problem in improving execution
concurrency for long-duration database transactions.

[科資中心編號]RA8510-2049
[題    名]
資料庫服務與電腦網路應用四年計畫(III)
Research and Development Project of Database Services and Computer
Network Application. (III)
[研 究 者 ]
王建敏
WANG,JIANN-MIN
[機構名稱  ]資訊工業策進會(IIIG)
[經費來源  ]經濟部(MOEA)
[補助金額  ]227633000
[補助編號  ]84-EC-2-A-15-0129
[研究開始日期]1994/07
[研究結束日期]1995/06
[出版日期  ]1995/06
[語    文]中文
[建檔日期  ]1997/05
[出版情形  ]研究報告,民國84年6月
Report, Jun. 1995
[科資分類號 ]IA0304;IA0411
[關 鍵 詞 ]
分散式資料庫;多媒體;交易處理;網路管理;辦
公室自動化;生產排程;異質資料庫

Distributed database;Multimedia;Transaction processing;Network
management;Office automation (OA);Production scheduling;
Heterogeneous database

[摘    要]
1.分散式資料庫存取管制模組:本系統整合
網路上分散的多個異質性資料庫,允許使用者
經由單一指令同時存取網路上多個資料庫系統
之內容,每一單指令均依資料分散情形,分解、
平行的分送至個點擷取所需資料。藉由資料存
取路徑最佳化、平行處理,網路多筆傳輸等處
理,資料存取速度優於異質性資料庫連結軟體

2.商用傳呼工作站:本產品提供電腦電話整
合應用之商務辦公室自動化服務,包括電話語
音查詢、傳真回覆、電話訂貨與調查、電話收
費等應用需求。更具備了系統快速操作設定與
多組樣本系統客製服務,可提升商用系統生產
力及操作效率。
3.多媒體電子書開發工具:本工具遵循SGML
多媒體文件標準及HYTIME超媒體鏈結標準,以及
提供中文處理、Hot Spot編輯、Style Sheet編輯等
能力。本系統操作簡易、符合多媒體文件標
準,滿足各類型多媒體電子書編輯之需求。

1. The goal of this system is to provide a standard interface
with high transparency of networked heterogeneous DBMSs for the
applications. The applications view the heterogeneous databases as a
centralized global database without knowing any detail about the
data distribution. The implementation of query decomposition,
parallel processing, network blocking fetch presents a well
performance for the system.
2. This product provides a computer/telephony integration
solution to the office automation needs, which include inquiry
service, FAX answering, ordering service. telephone polling, and
bill collection applications. with added features like quick
operating procedure setup and system customization options,
operating efficiency and business productivity will be greatly
improved.
3. The tool follows the multimedia document standard SGML
Format and hyperlink standard HYTIME format in addition, it also
provides the Chinese-Character processing capability, Hot Spot
editor, style sheet editor, and so on. The user-friendly interface
and standardized format can satisfy the need of all kinds of
electronic book developers.

[科資中心編號]RA8604-2045
[題    名]
聯合資料庫系統上一種新的查詢處理方法
A New Query Processing Strategy for Federated Database System.
[研 究 者 ]
陳耀輝;林金城
CHEN,YAW-HUEI;LIN,JIN-CHERNG
[機構名稱  ]大同工學院資訊工程研究所(TTCTINSG)
[經費來源  ]行政院國家科學委員會(NSC)
[補助金額  ]409000
[補助編號  ]NSC84-2213-E036-020
[研究開始日期]1994/08
[研究結束日期]1995/07
[頁 冊 數 ]30頁
[語    文]中文
[建檔日期  ]1997/11
[出版情形  ]完成日期:民國85年1月
[原件提供  ]行政院國家科學委員會科學技術資料中心
NSC84-2213-E036-020
[科資分類號 ]IA0304
[關 鍵 詞 ]
異質資料庫;查詢處理;聯合資料庫系統;查詢語言;物件導向資料庫

Heterogeneous database;Query processing;Federated database system;
Query language;Object-oriented database

[摘    要]
管理多個不同的獨立資料庫,使得使用者能在異質且分散的環境下存取資訊,已經變成愈來愈重要的課題了。相同的實體(Entity)可能會因不同的資料庫而有不同的型式,而且也可能被複製到多個地方去。關於這個問題,我們提出一個兩階段查詢處理策略,來決定查詢時重複資料取得的位置(Site)。在異質資料庫中,查詢處理的方式是將全域性的資料型態及語言對映成各別區域性的資料型態及語言,也就是將全域性的查詢語言轉變成多個異質的區域性查詢語言,並且讓使用者能夠透過全域性資料庫管理系統來存取異質資料庫。而我們假設這全域性資料庫管理系統是一個物件導向資料庫管理系統。在查詢處理策略的第一階段,我們使用以樣本為基礎(Pattern-base)查詢處理技巧來取得符合的查詢條件的物件;接著在第二階段,我們選擇最佳的位置(Site)來存取資料。在第一階段中,我們使用一些例子說明我們的啟發式(Heuristic)法則,然後在第二階段裡,我們做更進一步的證明。

It is becoming increasingly important to manage pre-existing
autonomous data-bases as a repository resource, so that users can
access information from such a heterogeneous distributed environment.
However, since the same entity in the real world may be modeled by
different pre-existing databases, some data are replicated at
different sites. In this research, we propose a two-phase query
processing strategy to determine the sites from which the query
retrieves the redundant data. A global-data-model-and-language to
local-data-models-and-languages mapping is adopted to process queries
in the heterogeneous databases. This mapping translates a query in
the global query language into several subqueries in different local
query languages, and thus allows the user to view and access local
heterogeneous databases via the global database management system,
which is assumed to be an object-oriented database management system.
In the first phase of the query processing strategy, we use a pattern-
based query processing technique to identify objects that satisfy the
specified global query. Then, in the second phase, we select the best
sites to retrieve the data. We use examples to illustrate the
heuristic rules used in the first phase, and provide a proof of
correctness of the second phase.

[科資中心編號]RB8702-0006
[題    名]
資料庫服務與電腦網路應用四年計畫(IV)
Database Service and Computer Network Application Technology R&D
Project (IV)
[研 究 者 ]
王建敏
WANG, MINA J.M.
[機構名稱  ]資訊工業策進會(IIIG)
[經費來源  ]經濟部(MOEA)
[補助金額  ]231475000
[補助編號  ]85-EC-2-A-15-0129
[研究開始日期]1995/07
[研究結束日期]1996/06
[語    文]中文
[建檔日期  ]1998/12
[出版情形  ]完成日期:民國85年6月
[科資分類號 ]IA0410
[關 鍵 詞 ]
異質資料庫;網路管理;電腦電話整合;排程;類神經網路;安全性

Heterogeneous database;Network management;Computer telephony
intrgration (CTI);Scheduling;Neural network;Security

[摘    要]
本計畫已完成異質性分散式資料庫系統,TCP/IP發展工具,SNMP網路管理系統,分散式應用系統管理工具,工作流程管理系統,多媒體電子書管理及開發工具,商用傳呼(CTI)系統,製造業廠區資訊系統,生產排程系統及及時供料資訊系統,建立起Client-Server、分散式處理、分散式資料庫、網路連線、網路管理、系統安全、多媒體等技術能力。其中Schema整合技術、ODBC連線技術、DBR關鍵資源生產排程等多項技術具有可與世界領先產品評比之地位。以上成果獲致65家次技術移轉案,技轉收入約1900萬元。其中多項成果已由廠商產品化銷售,而製造業廠區資訊系統及生產排程系統經實驗廠實際上線驗證後,可節省人力及提高產能,其效益達總營業額之10%,以國內製造業在整體經濟所佔之比重,其實際效益非常龐大。本計畫在網際網路連通技術之研發及成果之推廣,促使國內Internet商業應用之快速成長(Hinet及Seednet已達十萬多用戶),及促進網路應用之普及,提高資訊流通及使用之效率,提高全體社會之生產力及生
活方便性。

Research Accomplishment of FY 1996
This Database Services and Computer Network Applications
Research and Development Project has successfully completed planning,
design and implementation of (1) a Heterogeneous Distributed Database
System, (2) a TCP/IP Protocol Development Toolkit, (3) an SNMP
Network Management System, (4) a Distributed Application System
Management Toolkit, (5) a Workflow System, (6) a Multimedia
Electronic Book Development and Management System, (7) a CTI System
and Toolkit, (8) a Manufacturing Shop Floor Information System, (9) a
Shop Floor Scheduling System, and (10) a Just-In-Time Material Supply
Information System.
The Project enabled the establishment of strong capabilities in
Distributed processing, Client-Server, Distributed Database, Network
Connectivity, Network Management, System Security and Multimedia
technologies. In particular in those of Database Schema Integration,
ODBC Connectivity, DBR Critical Resource Scheduling areas, the level
of technology capability is comparable with, if not excel, that are
implemented in the world leading products.
Under this Project, the achievement includes 66 technology
transfer items to Taiwan software vendors with transfer-income of
NT$19 million. A large number of the transferred technological items
have been commercialized into products and made it available by the
receiving companies. Notable ones include that deployment of the
Manufacturing Shop Floor Information System and the Shop Floor
Scheduling System has achieved manpower reduction and productivity
improvement upto 10% of the company's total revenue as proven by
experimentations. As the proportion of Manufacturing sector weighted
high to the overall national economy, the potential benefit impact
can be phenomenal.
The R&D accomplishment, paired with the promotion of TCP/IP
connectivity, has accelerated the speedy growth of domestic Internet
services (subscribers to HiNet and SeedNet have reached more than one-
hundred thousand as of today). It has also motivated network usage
and popularity. Thus resulted in efficient information flow,
productivity improvement and universal access to the people.

[科資中心編號]RB9001-0023
[題    名]
資料庫服務與電腦網路應用四年計畫(IV)
N/A
[研 究 者 ]
王建敏
Wang, Mina J.M.
[機構名稱  ]資訊工業策進會(IIIG)
[經費來源  ]經濟部(MOEA)
[補助金額  ]231475000
[補助編號  ]84-EC-2-A-15-0129
[研究開始日期]1995/07
[研究結束日期]1996/06
[頁 冊 數 ]69頁
[語    文]中文
[建檔日期  ]2001/05
[出版情形  ]完成日期:民國85年7月
[科資分類號 ]IA0304; IA0202; SM0404
[關 鍵 詞 ]
資料庫服務; 網路應用; 分散式資料庫; 電腦網路; 多媒體資訊系統; 網路管理; 排程; 異質性資料庫; 系統安全性

Database service; Network application; Distributed database;
Computer network; Multimedia information system; Network management;Scheduling; Heterogeneous database; System security

[摘    要]
年度目標達成情形:(1)研發可整合市面上主要品牌資料庫系統之異質性分散式資料庫系統技術,以解決資料整合困難之問題,促進資訊之流通及應用。(2)研發網路管理技術、提供網路設施管理、分散式應用系統管理及網路流量管理等完整之解決方案,降低網路應用之困難度及運作成本。(3)開發網際網路連通技術,以促進網路使用普及,提高資訊之使用效率。(4)研發網路式之多媒體資訊技術,以協助國內業者提昇應用多媒體技術之能力。(5)完成製造業設計、規劃、現場製造完整迴路上之製造業現場資訊流整合系統,以滿足國內製造業者所需具高度彈性之資訊技術。

[科資中心編號]RB8707-0217
[題    名]
一個達到異質性資料庫間互通能力的詢問句處理器
Interoperable Query Processing among Heterogeneous Databases
[研 究 者 ]
張雅惠
Chang, Ya-Hui
[機構名稱  ]靜宜大學資訊科學系(PRUMINF)
[經費來源  ]行政院國家科學委員會(NSC)
[補助金額  ]220000
[補助編號  ]NSC86-2213-E126-006
[研究開始日期]1996/08
[研究結束日期]1997/07
[頁 冊 數 ]4頁
[語    文]英文
[建檔日期  ]1999/10
[出版情形  ]完成日期:民國87年2月
[原件提供  ]行政院國家科學委員會科學技術資料中心
NSC86-2213-E126-006
[科資分類號 ]IA0304
[關 鍵 詞 ]
異質資料庫;查詢處理;互通能力

Heterogeneous database;Query process;Interoperability

[摘    要]
隨著資料庫軟體與通訊工業的蓬勃發展,很多系統已經可以透過網路將資訊無遠弗屆地傳送給眾多的使用者。但是,目前的技術,並未考慮到資料庫間模式與查詢語言的差異性,使得他們之間沒有辦法達到完全的資源共享。本計畫設計一個詢問句處理器,有效地擴充關聯式資料庫與物件導向資料庫間的互通能力。

Recent advances in database technologies and communication
technologies have led to the demand for interoperability of different
information servers. In this project, we have developed techniques
for producing interoperable queries with object and relational
databases. A user poses a local query in a local query language,
against a local object or relational schema. We transparently produce
appropriate queries with respect to remote target object or
relational databases, which contain data that are relevant to the
user, so that data can be shared. A set of heterogeneous
transformation algorithms define the appropriate transformations from
the local queries to the remote queries.


[科資中心編號]RB8910-0204
[題    名]
一致化資料庫的建構(I)
Developing a Reconciled Database (I)
[研 究 者 ]
陳煇煌; 黃士銘
Chen, Huei-Huang; Huang, Shi-Ming
[機構名稱  ]大同大學資訊工程系(TTUIE)
[經費來源  ]行政院國家科學委員會(NSC)
[補助金額  ]446000
[補助編號  ]NSC88-2213-E036-002
[研究開始日期]1998/08
[研究結束日期]1999/07
[語    文]中文
[建檔日期  ]2001/06
[出版情形  ]完成日期:民國88年12月
[原件提供  ]行政院國家科學委員會科學技術資料中心
NSC88-2213-E036-002
[科資分類號 ]IA0304
[關 鍵 詞 ]
資料一致性; 資料庫綱要整合; 多資料庫; 異質性資料庫; 資料倉儲; 綱要轉換; 綱要整合; 逆向工程; 處理模型

Data consistency; Dababase schema integration; Multidatabase;
Heterogeneous database; Data warehouse; Schema translation; Schema
integration; Reverse engineering; Process model

[摘    要]
在過去的幾十年間,已經有相當多的研究討論資料庫綱要整合的問題。儘管完整的整合綱要一直是這個研究的目標,但是卻很少有令人滿意的方法被提出來,尤其是關於資料庫綱要的比對。主要的原因在於一個概念可以被不同的人用不同的方式表現出來。在反向工程裡,當我們在整合數個資料庫綱要時,這將會造成許多語意上的混淆和產生出一些嚴重的問題。而在以往已經提出來的方法中,解決這些語意上不清楚的問題大多是依賴資料庫的管理者來提供相關領域的知識以協助整合。這也就是說人在整合的過程中扮演了很重要的角色,並且要肩負起整合結果的成敗責任。正如我們上面所說的,要發展一套全自動的資料庫綱要整合的方法是不可行的,卻可以找出一些方法來幫助使用者找出一些語意或是確認一些語意,並且減少對人的依賴。在本研究中,我們認為對處理模型加以處理是提供我們更多有關綱要整合語意一個可行的方式。

Over the past decades, a considerable number of studies have
been devoted to database schema integration. Although completely
integrated schema has been an objective of study for a long time,
only a few satisfying methodologies have been proposed, especially
about schema comparison. The primary cause is that a concept can be
modeled in many different ways by different people. In reverse
engineering, it would cause several semantic confusions and many
critical problems when we were integrating several schemas. Most of
the resolutions of these issues in proposed methodologies are
dependant on database administrators (DBAs) to provide some domain
knowledge to aid the process of integration. That is to say that
human plays an important role in integrating phase and saddles
responsibility with the success or failure about integrated results.
It would, as we said above, be ineffective to develop an automatic
methodology for schema integration, but it is possible to find some
methods to assist users in discovering or checking correspondences,
and reduce dependence on human. In this research, we consider that
taking process models into account is a practicable way to provide
the integration more semantics and decrease the reliance on the
DBAs.

[科資中心編號]RB8910-0206
[題    名]
網際網路異質性資料庫資源整合之研製
Implementation of a WWW Heterogeneous Database Server
[研 究 者 ]
曾秋蓉
Tseng, Judy C.R.
[機構名稱  ]中華大學資訊工程研究所(CHUINFEG)
[經費來源  ]行政院國家科學委員會(NSC)
[補助金額  ]458000
[補助編號  ]NSC88-2213-E216-020
[研究開始日期]1998/08
[研究結束日期]1999/07
[語    文]中文
[建檔日期  ]2001/06
[出版情形  ]完成日期:民國88年12月
[原件提供  ]行政院國家科學委員會科學技術資料中心
NSC88-2213-E216-020
[科資分類號 ]IA0304; IA0202
[關 鍵 詞 ]
異質性資料庫; 全球資訊網; 資訊檢索; 網站伺服器; 網際網路; 搜尋引擎

Heterogeneous database; World Wide Web (WWW); Information retrieval;Web server; Internet; Search engine

[摘    要]
隨著全球資訊網(WWW)的蓬勃發展,在網際網路上提供各式各樣資訊的網站也越來越多,這些網站各自管理著自己的資料庫系統,以提供資訊給上站查詢的使用者。然而由於這些資料庫系統之間並無關連,因此使用者必須一一造訪各個網站,並且自行整理所獲得之資料,才能得到統整的資訊。同時在這些資料庫系統中所存在的大量重覆性的資料,也會隨著使用者的查詢一而再,再而三的重覆出現。如此一來,更增加了使用者在搜尋有效資訊時的困難度。目前WWW上的資訊網站,絕大多數都是在一個網站伺服器(Web Server)上,用網站發展者所熟悉的資料庫管理系統,去維護所欲提供的資訊。當用戶端(Client)的使用者對此網站發出查詢的要求時,伺服器即透過所採用的資料庫管理系統,從資料庫中搜尋出使用者所想要的資訊傳回給使用者。在這種架構下,使用者拜訪一個網站只能取得存在此網站資料庫中的資料,如果查不到所要的資訊,則使用者必須自行鍵入其他同類型網站的網址(Uniform Resource Locator,URL),切換到其他網站後,再下達一次同樣的查詢,以取得存在其他網站資料庫中的資訊。這個過程必須不斷地重覆直到搜尋到足夠的資訊為止,非常沒有效率。
雖然目前有不少網站號稱可以整合網際網路上的資訊,如搜尋引擎
(Yahoo、蕃薯藤、GAIS、SavvySearch)、購物網站(AcerMall、Amazon)...等,但這些網站大部份只是運用了網頁嵌入的技術,將相關網站的網址記錄在網頁中,供使用者點選以納入其他網站的資訊。使用者雖然少掉了記憶網址的負擔, 卻仍然需要重覆下達相同的查詢以及自行過濾重覆的資料,查詢過程仍然十分繁複。為了能夠依據使用者的需求去各個網站查詢資訊,加以統整後再一次呈現給使用者,於是便有了匯總式搜尋引擎這類的搜尋代理程式(Agent)產生。代理程式除了能夠提供使用者下達命令或要求外,還能將多個網站所提供的資料加以匯整、過濾、比對並刪除重覆的資料後,再回傳給使用者做參考或利用,讓使用者能夠在最短的時間內找到想要的資訊。由於網際網路上這類代理程式的應用非常多,除了最常用的搜尋引擎之外,日漸興起的網路購物也是代理程式的應用之一,而代理程式最重要的工作便是克服各網站內部資料庫的異質性,使資料能夠加以整合,因此本研究計劃實作了一個網際網路異質性資料庫伺服站(WWW Heterogeneous Database Server, W3HDS)。此伺服站可協助代理程式搜集來自各網站資料庫的資訊,消除其資料的異質性,並加以匯整及過濾,提供整合的資訊給代理程式做進一步的處理。因此本計畫的完成將使得網際網路代理程式的開發更加容易,無論將之應用於搜尋引擎或是網路購物上,都能使網際網路所提供的大量資料不再成為使用者的負擔,而能真正發揮網際網路資訊無遠弗屆的好處。

With the growth of the World Wide Web (WWW), there are more
and more Web Servers established on the Internet. Each Web Server
maintains a database of its own to provide information to web users.
Search engines and catalogue systems can do an excellent job of
finding relevant sites or pages for users. However, these search
engines still have some problems. First of all, since there is no
connection among these web databases (or search engines), users
have to visit the web servers one by one to obtain the information
needed. It is difficult for a user to remember all the URL's of the
web servers to be visited. Moreover, it is also awful for a user to
read the large amount of repeating data. In this project, we will implement a WWW Heterogeneous Database Server (W3HDS) to integrate the databases distributed in the Internet. When a WWW user submit a request to W3HDS, W3HDS will broadcast the request to associated Web Servers. After receiving the result return from the web servers, W3HDS will integrate all the results, remove duplicated data, and then return back to the user. With W3HDS, users only submit requests to a single web server
(the W3HDS), hereafter they obtain information from multiple web
servers. This project perfectly combines the research areas of WWW and
Heterogeneous databases, and also makes WWW more applicable and
more convenient for users to obtain Internet information.

[科資中心編號]RB8910-0212
[題    名]
在異質性資料庫上之多維查詢法則
Multi-Dimension Query in Heterogeneous Database
[研 究 者 ]
黃士銘; 羅偉丞; 彭錦淮
Huang, Shi-Ming; Lour, Weei-Cherng; Perng, Jiin-Hwai
[機構名稱  ]大同大學資訊工程研究所(TTUINFEG)
[經費來源  ]行政院國家科學委員會(NSC)
[補助金額  ]324000
[補助編號  ]NSC88-2213-E036-005
[研究開始日期]1998/08
[研究結束日期]1999/07
[語    文]中文
[建檔日期  ]2001/06
[出版情形  ]完成日期:民國88年12月
[原件提供  ]行政院國家科學委員會科學技術資料中心
NSC88-2213-E036-005
[科資分類號 ]IA0304
[關 鍵 詞 ]
異質性資料庫; 多維度查詢; 資料倉儲; 決策支援系統

Heterogeneous database; Multidimension query; Data warehouse;
Decision support system

[摘    要]
現今的資料庫系統查詢語言,大多只提供一維或二維的資料查詢法,例如:SQL、OSQL等等,這種類型的資料查詢法僅適用於傳統資料處理的表單作業方式,當資料要提升為資訊甚至於智慧以提供給經理人員下決策作參考時,則不足以適用,例如:某銷售部門經理,對於同一季的銷售資料需由通路、地區、產品類別、年度來同時查看,才可以比較精確的下決策,因而多維查詢就益加顯得必須與重要。傳統的作法在解決多維查詢的問題,往往須要撰寫非常複雜的程式來達到此目的,因而若要變更多維查詢的條件,則就變成一件非常複雜的工作。但是多維查詢的利用往往是在一很動態的環境之下,如決策支援、智慧型的查詢等。因而需有一彈性化且高效能的多維查詢法則,傳統的方法已不能適用。
本計劃主要的目的是要將目前資料庫查詢的方式做一有效的改善,其方法是利用多維度查詢來改善傳統的單維度或是二維度的簡易查詢,以期能在現存的資料庫中查詢出更多有用的資訊,以利決策支援系統的開發。目前在各企業間的資料庫系統,其架構往往不盡相同,例如關聯式資料庫、物件導向式資料庫等等,其間的查詢方式都不同,所以有許多相關的資料都分散在不同的系統中,使得這些相關的資料都無法得到妥善的利用,如此無形中造成了資料及資訊的浪費,所以我們也需要找出一整合的方式,來統合整理這些不同的資料庫系統。

Data warehousing or decision support applications need to
aggregate data across many dimensions looking for unusual useful
patterns. Most of current database languages, such as SQL, only
allow users to retrieve the data in zero-dimensional,
one-dimensional, or two-dimension aggregates. Applications need the
N-dimensional generalization of these operators. Unfortunately
current solutions need to require the application programmers to
develop complex programs for these requirements. They are
inflexible and can not be dynamically modified. These cause the
applications unfriendly or even failure. Furthermore, most of
current multi-dimension query system only can support on one DBMS,
but a lot of real applications are in the heterogeneous environment.
In this project, we describe an architecture which can allow users
to do directly multi-dimension query in the field-based integration
heterogeneous database system.

[科資中心編號]RB9001-0022
[題    名]
使用行動代理者作分散式資料庫資料擷取
Using Mobile Agents for Distributed Database Access
[研 究 者 ]
王宗一
Wang, Tzone-I
[機構名稱  ]成功大學工程科學系(NCKEENS)
[經費來源  ]行政院國家科學委員會(NSC)
[補助金額  ]710000
[補助編號  ]NSC89-2213-E006-056
[研究開始日期]1999/08
[研究結束日期]2000/07
[頁 冊 數 ]5頁
[語    文]中文
[建檔日期  ]2001/05
[出版情形  ]完成日期:民國89年12月
[原件提供  ]行政院國家科學委員會科學技術資料中心
NSC89-2213-E006-056
[科資分類號 ]IA0304
[關 鍵 詞 ]
行動代理人; 分散式交易處理; 分散式異質資料庫; 交易管理; 主從式架構;
鬆耦合性分散式系統

Mobile agent; Distributed transaction processing; Distributed
heterogeneous database; Transaction management; Client server
architecture; Loosely coupled distributed system

[摘    要]
本計畫主要是從如何把一分散式資料庫之交易管理(Transaction
Management)架構在非主從計算(Client/Server Computing)方式上之研究角度來探討及建立一行動代理者組織架構(Mobile Agent Framework),以支援分散式異質資料庫管理系統在不特定網路(或網際網路)上對鬆散式整合(Loosely coupled)之分散式資料庫作資料擷取。

The main purpose of this project is, from a research
perspective of not establishing the transaction management of a
distributed database management system on a client-server
architecture, to use a Mobile Agent Framework for supporting
effective access to a distributed heterogeneous database integrated
with loosely coupled local databases scattered in a network.

[科資中心編號]RB8901-0519
[題    名]
虛擬半導體廠之異質資料庫資訊伺服器(I)
Virtual Factory Information Server in Heterogeneous Database
Environments (I)
[研 究 者 ]
許永真
Hsu, Yung-Jen
[機構名稱  ]台灣大學資訊工程系(NTUECSIE)
[經費來源  ]行政院國家科學委員會(NSC)
[補助金額  ]513000
[補助編號  ]NSC88-2218-E002-012
[研究開始日期]1998/08
[研究結束日期]1999/07
[頁 冊 數 ]5頁
[語    文]中文
[建檔日期  ]2000/06
[出版情形  ]完成日期:民國88年7月
[原件提供  ]行政院國家科學委員會科學技術資料中心
NSC88-2218-E002-012
[科資分類號 ]IE0602; IE0507
[關 鍵 詞 ]
異質環境; 半導體製造; 虛擬工廠

Heterogeneous environment; Semiconductor fabrication; Virtual
factory

[摘    要]
本計畫的目標是在設計一個虛擬的自動化半導體工廠,以使客戶可充分控制產品進度,及時獲得製造流程上的正確資訊。
整個自動化製程中,資料源往往由數個專屬系統分散管理,例如,「全方位定單管理系統」、「製造執行系統」如Workstream或Promis、「工程資料分析系統」如SAS、以及「定料系統」(Bill of materials)等。然而,這類現行系統往往屬封閉式架構,各系統均有其專屬之資料格式與查詢方式,使得資料之交換與整合不易,本子計畫之目的便在於整合虛擬半導體廠計畫中的所有資料源,及應付不同型態的資料、不同目的的存取,因而建構一整合性的異質資訊伺服器(Information server)。

This project aims to construct an information server that
powers a "virtual factory". The information server provides the
core functionality of a data integration virtual factory. A user
can query the server without regard to the specific storage
mechanism and/or generation process for the individual pieces of
information. The primary research issues involved include:
information server architecture for distributed, heterogeneous
database environments, correspondence of the existing data with a
relational database, techniques for query planning, data mapping,
and operation mapping, just-in-time server technology for real-time
data access, and web-enabled query interface to distributed,
heterogeneous databases.

[科資中心編號]RB8707-0263
[題    名]
異質性多資料庫環境總體交易管理之研究
A Study on the Global Transaction Management in Heterogeneous
Multidatabase Environment
[研 究 者 ]
楊鍵樵;連志誠
Yang, Chen-Chau;Lien, Chih-Cheng
[機構名稱  ]台灣科技大學電子工程系(NTSTENT)
[經費來源  ]行政院國家科學委員會(NSC)
[補助金額  ]574300
[補助編號  ]NSC85-2213-E011-025
[研究開始日期]1995/08
[研究結束日期]1996/07
[頁 冊 數 ]126頁
[語    文]中文
[建檔日期  ]1999/10
[出版情形  ]完成日期:民國85年7月
[原件提供  ]行政院國家科學委員會科學技術資料中心
NSC85-2213-E011-025
[科資分類號 ]IA0304
[關 鍵 詞 ]
多資料庫系統;交易管理;異質性;順序化;單一性;死結

Multidatabase system;Transaction management;Heterogeneous;
Serializaility;Atomicity;Deadlock

[摘    要]
多資料庫系統是一個整合多個異質性的本地資料庫設施,並容許使用者存取各個自主行性本地資料庫中的資料。本研究計畫旨在探討多資料庫環境下總體交易管理的問題,包括了總體交易的衝突偵測、總體順序化的達成、交易失敗之回復、複本資料管理及總體死結偵測策略等等。為達成多資料庫交易總體順序化,我們提出有向一致圖理論,以決定交易的執行時機,使其避免因本地父易造成總體子交易執行次序與順序化次序不一致。此外,在交易失敗的環境下,透過二階層確認協定及針對交易回復時的特性,對有向一致圖增加適當的狀態,以確保總體順序化仍可以維持。複本資料的管理,以各個本地伺服器間可互相協調及維護有向一致圖,以決定交易的次序與執行時機,使得本地站與總體系統對複本資料更新皆在一致的情況下進行,達成一份拷貝總體順序化的目標。另外,總體死結偵測方面,由於本地自主性的關係而使得系統無法得知本地等待圖,故引用潛在衝突圖來替代之。並再結合總體等待圖與等待確認圖等來監控總體交易的執行狀況,以作為判斷死結發生與否之依據進而維護系統之運作。

A multidatabase system is a facility that integrates multiple
autonomous, heterogeneous database management systems (DBMS) and
allows users to access data located in each of the DBMS. The purpose
of this research project is to explore the problems of global
transaction management in a multidatabase environment. It includes
that conflict detecting of global transactions, the ensuring of
serializability, the recovery of transaction failure, the management
of replica data, and the detection of global deadlock etc. To achieve
global serialization, we proposed the directed consistency graph
theorem then use it to decide when to submit transactions, thus it
can conduct an execution order which is consistent with serialization
order of global subtransactions. In a failure prone environment while
transaction recovers, the global serializable must be ensured by
using two phase commit protocol and expanding the directed
consistency graph theorem.
While replica data are updated, the replica data management
charges to keep one copy global serializable, by letting the local
servers be coordinated and maintaining a directed consistency graph.
About the global deadlock, the local sites are autonomous so the
system cannot obtain local wait-for graphs, thus we take the
potential conflict graphs to replace them. By integrating potential
conflict graph with global wait-for graph and wait-for commit graph,
they can be used to monitor the execution status of global
transactions in the local sites and to judge whether global deadlock
happens or not.

科資中心編號: RB9104-0090 
題名: 一致化資料庫的建構(3/3) 
Developing a Reconciled Database 
研究者: 陳煇煌 ; 黃士銘 
Chen, Huei-Huang ; Huang, Shi-Ming 
機構名稱: 大同工學院資訊工程系 
經費來源: 行政院國家科學委員會 
補助金額: 新台幣 434 仟元 
補助編號: NSC89-2213-E036-048 
研究日期: 2000/08 - 2001/10 
出版日期: 2001/08 
頁冊數: 6頁 
原件提供: 行政院國家科學委員會科學技術資料中心, NSC89-2213-E036-048 
科資分類號: IA0304 
關鍵詞: 一致化資料庫 ;擴充式實體關聯模式 ;資訊整合 ;可擴展標示語言 ;綱要轉換 
Reconciled Database ;Enhanced Entity Relationship Model (Eer Model) ;Information Integration ;Extensible Markup Language (Xml) ;Schema Translation 
中文摘要: 現今資料庫的研究中,對於解決異質性資料庫系統的問題,多採用"多資料庫系統"的觀念來解決其相容性的問題。我們已建立異質性資料庫間語法的互相轉換與資料同步技術。在此基礎之上,本計劃利用這些技術來建立一個一致化的資料庫系統環境,也就是資料一致、資訊一致、以及系統一致的環境,以解決多資料庫系統中資訊整合與一致性程度偏弱的問題。本計畫第三年的目的,在於提出一個一致化的機制,包含了面對不同型態資訊的分析方法論,將網際網路上不同型態的分散式資訊來源模式架構:包括資料庫應用程式與網頁資訊來源等等,進行一致化的動作,統一以具資料庫概念之擴充式實體關聯(EER)資料模式表示,並利用預先建立之範圍規則(Rules)預測其資訊語意;各資料來源的資料模式並透過衝突的解決與合併,利用拼圖法整合成一個以依據使用者需求所定義的資料模式藍圖為基礎,並針對各資料來源均通用的資料模式語意,以達到將 Web-based 分散式資料來源上資訊語意整合的目標,以期讓使用者與應用程式都能以一致的表示方式,用更具廣度與更方便的方式,呈現、了解、查詢,並利用分散式Web 資料來源上的資訊。 
英文摘要: The concept of Multi-database system has been used on heterogeneous database environment. However, the data integrity and consistency is a problem. The reconciled database system in our research is an improvement. It represents a solution to the problems by offering an architecture of a reconciled database for the connectivity among information resource, including traditional DBMS and modern Web information resources, etc., with different data models. In our study, a mechanism for Web information integration by using Enhanced Entity Relationship (EER) Model is investigated. With a universal view of different heterogeneous information, after our integration process users in enterprises can therefore understand and analyze Web information in a more convenient and broad way. 

科資中心編號: RS9009-0036 
題名: 多資料庫系統整合處理及管理之研究(II) 
Stydy of Multidatabase System's Integration, Processing and Management(II) 
研究者: 楊鍵樵 
Yang, Chen-Chau 
機構名稱: 台灣科技大學電子工程技術系(四年制) 
經費來源: 行政院國家科學委員會 
補助金額: 新台幣 424 仟元 
補助編號: NSC83-0408-E011-010 
研究日期: 1994/02 - 1995/01 
出版日期: 1995/07 
頁冊數: 117頁 
原件提供: 行政院國家科學委員會科學技術資料中心, NSC83-0408-E011-010 
科資分類號: IA0304 
關鍵詞: 多資料庫系統 ;查詢最佳化 ;網目整合 ;一致圖理論 ;共同資料模式 ;推導式資料庫系統 
Multidatabase System ;Query Optimization ;Schema Integration ;Consistency Graph Theory ;Common Data Model ;Deductive Database System 
中文摘要: 本文是'多資料庫系統整合、處理及管理之研究'計畫的第二期成果報告,內容包括:共同資料模式的選擇、查詢處理、綱要整合、資料庫維護、以及資料庫交易管理。多資料庫系統的目的,在於使用一個統一的介面,來有效地存取、管理儲存在各本地資料庫中的資料。因本地資料庫可能為異質的,因此需要一個共同的資料模式來表示各資料庫綱要。我們選擇邏輯資料模式為本系統的共同資料模式,因其語意豐富,且與關聯式資料庫有相同的理論基礎。為了模式化各種資料模式,我們擴充其為具有階層化、特殊化、聚集、及抽象資料型態的能力,而成為物件導向邏輯資料模式,並提出一個半自動的綱要整合方法。更新為破壞資料庫一致性的原因之一,為了保證資料庫的完整性,我們發展了一個自動化更新傳播機制,以使資料更新時能同時更新其他相關的資料。針對多資料庫交易總體順序化問題,文中提出一個三階層並行控制架構,運用我們所提出的一致圖理論,以偵測衝突方法,決定交易的執行時機,以避免本地交易造成總體子交易執行次序與順序化次序不一致。在此一系統架構下,我們並證實其確可達成總體順序化。 
英文摘要: In this report, we present the research results of the second year of the project 'A study of multidatabase system's integration, processing and management.' The selection of common data model, query processing, schema integration, database maintenance, and transaction management are discussed. A multidatabase system is a system that integrates the operational data of several autonomous database systems and provides a uniform interface and control mechanisms to control access to those data. Because the participating local databases may be heterogeneous, we need a common data model to describe the database schemas of the local databases. In this thesis, we select the logic data model as the common data model for its rich semantics, and having the same underlying theory with the relational data model. For model a variety of data models, we extend logic data model to object-oriented logic data model with the generalization, specialization, aggregation, and abstract data type capabilities . In this report, we discuss the difficulties of schemas integration owing to data heterogeneity and present a semi-automatic approach to schema integration. Queries present no particular problems, while update may cause inconsistencies among databases. For database update integrity, we develop an automatic update propagation mechanism to update the relate data. Global serializability and local autonomy are the important issues in multidatabase systems. We proposed a three-level concurrency control architecture to resolve the above issues. We developed a consistency graph theory to detect conflicts and decide the submission of transactions so that the execution order and the serialization order of global subtransactions are consistent. The correctness of global serializaility of this system architecture is proved. 


科資中心編號: RB9101-1014 
題名: 主動式異質資料庫系統之資料一致性維護---應用於電子商務
The Maintenance of Data Consistency in Active Heterogeneous Database Systems---An Electronic Commerce Application 
研究者: 簡永仁 
Jean, Yeong-Ren 
機構名稱: 靜宜大學資訊管理系 
經費來源: 行政院國家科學委員會 
補助金額: 新台幣 301 仟元 
補助編號: NSC89-2213-E126-021 
研究日期: 2000/08 - 2001/07 
出版日期: 2001/10 
頁冊數: 4頁 
原件提供: 行政院國家科學委員會科學技術資料中心, NSC89-2213-E126-021 
科資分類號: IA0304;IA0202 
關鍵詞: 電子商務 ;異質資料庫 ;主動式異質資料庫系統 ;資料一致性 ;主動規範 
Electronic Commerce ;Heterogeneous Database ;Active Heterogeneous Database System (Ahdbs) ;Data Consistency ;Active Rule 
中文摘要: 電子商務勢將成為二十一世紀最主要的商業運作模式,其欲成功所不可或缺的必要條件包括了安全、有效率的物流、金流與資訊流,而這些必要條件要能夠充分成立,勢必仰賴一套優質的資料庫系統解決方案的後備支援。如今由於電子商務的需求,在整個電子交易過程中勢必牽連到買方、賣方、銀行等不同單位的不同型態資料庫系統,因此電子交易系統進行資料處理時,如何在內含不同型態資料庫的異質資料庫系統(Heterogeneous DatabaseSystem)中確保「資料一致性」(dataconsistency)已成為「電子商務」中的重要研究課題。在本計畫中,我們以現今已有的事件驅動式(event-driven )主動資料庫規範(active database rules )為基礎,提供更豐富的模式(coupling model )設定,使資料庫管理者有更多的彈性來設定當某些可能危害系統「資料一致性」的事件發生時的因應之道,以確保在電子交易應用中,異質資料庫系統的「資料一致性」。同時我們也提出一套更有效率的規範抵觸解決方案(rule-conflict resolution ),以預先且有效率的方法解決所設定的不同主動規範間相互抵觸的問題。另外我們也以此主動式異質資料庫系統(AHDBS, ActiveHeterogeneous Database System)為基礎,建立一套簡易的虛擬模型,藉以測試證實我們的方法機制可確實、有效率地維護「異質資料庫資料的一致性」。 
英文摘要: An electronic commerce application is concerned with various databases that are based on various database models and belong to various database systems. Therefore, heterogeneous database systems are very applicable to the electronic commerce application. Because of this informal and complex business environment in the real world, the data is very prone to inconsistency. Hence, the maintenance of data consistency among the various local database systems participated in the global heterogeneous database system is very important and complicated. Different from the passive strategy in the traditional database systems, Active Heterogeneous Database System (AHDBS) use the active rule mechanism to provide a more active and automatic solution of maintaining data consistency. The active rule is an event-driven mechanism that automatically monitors the pre-specified condition of the database state and takes some corresponding action when the condition comes into existence. However, the flexibility of the coupling model and the efficiency of the rule conflict resolution are both not fully satisfied our requirements today. In this project, we extend the coupling model of current active rule mechanism and improve the rule conflict resolution to provide a more dynamic, more flexible, and more efficient solution to the maintenance of data consistency in active heterogeneous database systems. And, we construct a prototype system of a simulated electronic commerce application to prove our improved active rule mechanism is practicable and valuable. 

科資中心編號: RF9101-0471 
題名: 以XML做異質性資料庫整合以建置資料倉儲之研究 
A Study on Heterogeneous Database Integration Through XML for Data Warehouse Creation 
研究者: 曾守正 
Tzeng, Shoou-Jeng 
機構名稱: 高雄第一科技大學資訊管理系 
經費來源: 行政院國家科學委員會 
補助金額: 新台幣 396 仟元 
補助編號: NSC89-2416-H327-033 
研究日期: 2000/08 - 2001/07 
出版日期: 2001/09 
頁冊數: 5頁 
原件提供: 行政院國家科學委員會科學技術資料中心, NSC89-2416-H327-033 
科資分類號: IA0304 
關鍵詞: 資料倉儲 ;異質性資料庫 ;可延伸標示語言 ;資料庫整合 ;綱要整合 
Data Warehouse ;Heterogeneous Database ;Extensible Markup Language(Xml) ;Database Integration ;Schema Integration 
中文摘要: 資料倉儲在近幾年受到企業界普遍的重視,因為企業大多意識到:想要有現代化的決策制定品質,那麼就必須將企業本身所累積的商業資料做完整規劃後,以多維度的分析來呈現方能達成。然而,資料倉儲的建立有賴整體企業與外部資訊的資料庫整合,整個過程牽涉到的異質性整合問題相當複雜,使得許多參與整合人員的共同結論無非是:需要花費大量人工,因此容易造成錯誤,此一觀念無形中也成了資料倉儲建置過程的絆腳石。但是,我們認為此整個問題並沒有如此悲觀,因為其解決方案與異質性資料庫整合的問題幾乎是一樣的,而此一研究在近二十年來已經被完整討論,具有相當完備的解決方案,而目前所欠缺的則是如何將這些研究與網際網路上的流通技術整合。因此,在本研究中我們提出了一個以 XML 為基礎的異質性資料庫整合架構,整合的標的當然是以資料倉儲為主。我們針對各式各樣的綱要整合問題提出了一個流暢的整合架構,它主要是以XML 為載體的整合模式與綱要整合問題解決方案。此一架構讓資料庫整合與資料倉儲的建置可以運用相同的架構與技術,降低兩者在建置過程中所遇到的阻力,並加速資料倉儲建置的速度。 
英文摘要: Data warehousing is gaining in popularity as organizations realize the benefits of being able to perform multi-dimensional analyses of cumulated historical business data to help contemporary administrative decision-making. However, since a data warehouse creation needs to integrate various enterprise-wide corporate data into a single repository, from which users can query via various dimensions and produce analysis report, there is still one key stumbling block to the development of data warehouses. Specifically, problems may arise in building and populating a data warehouse with existing data, since it has various types of heterogeneity. That leads to a general conclusion that the data warehouse creation task is usually labor-intensive, error-prone, and frustrating. Such misunderstandings lead to a number of data warehousing projects to be abandoned mid-way through development. In fact, the problems that are being encountered in data warehouse establishment are very similar to those encountered in heterogeneous database integration, which have been well studied for about two decades. That makes the situation seems to be not as tough as it appears. However, there are still works need to be accomplished to bring these achievements to be easily implemented and integrated to Internet applications. In this project, we review the general problems of heterogeneous database schema integration and intend to identify the common issues in data integration and data warehouse creation. We propose a platform for integrating heterogeneous data for data warehouse creation by employing XML formats in a seamless manner. The proposed methodology is not only suitable for heterogeneous database integration, but also suitable for data warehouse creation and web presentation. 

科資中心編號: RB9104-0260 
題名: 異質工作站叢集之快速集體式溝通 
Efficient Collective Communication in Heterogeneous Clusters 
研究者: 吳真貞 
Wu, Jan-Jan 
機構名稱: 中央研究院資訊科學研究所 
經費來源: 行政院國家科學委員會 
補助金額: 新台幣 318 仟元 
補助編號: NSC90-2213-E001-027 
研究日期: 2001/08 - 2002/07 
出版日期: 2001/07 
頁冊數: 4頁 
原件提供: 行政院國家科學委員會科學技術資料中心, NSC90-2213-E001-027 
科資分類號: IA0202 
關鍵詞: 異質性工作站叢集 ;群集通訊 ;群播 ;排程演算法 ;異質性網路 
Heterogeneous Workstation Cluster ;Collective Communication ;Multicast ;Scheduling Algorithm ;Heterogeneous Network 
中文摘要: 異質群組環境中之集體式溝通之探討乃是一個極新的領域,目前為止相關的文獻並不多。本計劃有三項主要貢獻:(1) 我們針對異質群組環境,提出一較精確之模式以估算資料傳輸之時間和效率。(2) 我們以此模式為基礎,設計集體式溝通的排序演算法。(3) 我們在CSIM之上架構一異質群組的模擬器,作為評估資料傳輸軟體在各種群組環境之執行效率 
英文摘要: In this proposed effort, we study the collective communication problem in HNOW systems. We propose a realistic communication model for heterogeneous systems. Based on the communication model, we investigate collective communication problems on HNOW systems and design efficient scheduling algorithms for them. We have also developed a CSIM-based network simulator for HNOW systems and the associated simulation packages as a test bed for evaluating collective communication algorithms on a variety of heterogeneous systems.

 科資中心編號: RN9101-0187 

題名: 應用影像處理技術於不同被膜處理鑽頭磨耗之自動量測

The Application of Image Processing Technique to the Automatic Wear Measurement of Different Coated Drills 

研究者: 陳文洲

Chen, Wen-Chou 

機構名稱: 大華技術學院自動化工程科 

經費來源: 行政院國家科學委員會 

補助編號: NSC89-2212-E233-007 

研究日期: 2000/08 - 2001/07 

出版日期: 2001/10 

頁冊數: 5頁 

原件提供: 行政院國家科學委員會科學技術資料中心, NSC89-2212-E233-007 

科資分類號: IA0403;IF0408;IF0502 

關鍵詞: 影像處理 ;刀具磨耗 ;塗層鑽頭 ;磨耗量測

Image Processing ;Tool Wear ;Coated Drill ;Wear Measurement 

中文摘要: 本研究主要目的是利用影像處理技術(Image processing technique)自動量測不同被膜處理鑽頭鑽腹面的磨耗量,提供線上(On line)監測鑽削加工時鑽頭之狀態,以取代利用工具顯微鏡離線(Off line)量測鑽頭腹面磨耗型態及磨耗量。同時,對自動量測鑽頭腹面磨耗量的系統架構、鑽頭磨耗檢測原理與方法等亦有詳細的介紹。 

英文摘要: The purpose of this research is to use the imageprocessing technique for the automatic wearmeasurement of different coated drills and on-linedetecting the drill state during the drilling processes inorder to obtain a better hole quality. The automaticmeasurement system, the principle and the method ofthe drill wear investigation are also depicted in thispaper. 

 

 

科資中心編號: RN9102-0119 

題名: 藉由基因遺傳演算法則達成模糊濾波器與任意複數係數FIR數位濾波器之設計應用

Genetic Algorithm Approach for Designing Fuzzy Filters and Arbitrary Complex Coefficient FIR Digital Filters 

研究者: 呂虹慶 

機構名稱: 大同工學院電機工程系 

經費來源: 行政院國家科學委員會 

補助編號: NSC89-2218-E036-008 

研究日期: 2000/08 - 2001/07 

出版日期: 2001/07 

頁冊數: 4頁 

原件提供: 行政院國家科學委員會科學技術資料中心, NSC89-2218-E036-008 

科資分類號: IE0305 

關鍵詞: 模糊濾波器 ;數位濾波器 ;遺傳演算法 ;加權函數

Fuzzy Filter ;Digital Filter ;Genetic Algorithm ;Weighting Function 

中文摘要: 在本計劃中我們將藉由基因遺傳演算法則設計一個全新的非線性模糊濾波器,將此濾波器應用於影像處理中,藉以剔除混合雜訊,例如,含有高斯雜訊與脈衝雜訊。我們利用區域統計的概念,並藉由所提出的基因遺傳演算法則透過適當的交配、突變與選取,演化出適當的歸屬函數值,將可以輕易的處理含有混合雜訊的影像,而且保留邊界的清晰度,不會因而模糊不清。實數基因遺傳演算法則可延伸至複數事例,以達成任意複數係數之有限脈衝響應數位濾波器設計。藉由通帶與截止帶所量測出來誤差平方使之最小化,如此便能使得染色體經由演化而獲得複數濾波器之係數,此種方法是利用遞迴複數基因遺傳演算法則,以便獲取與想要設計複數頻率響應的一個最小平方的近似值,在每一次遞迴中的加權函數則是使用上一次的遞迴結果予以更新,因此在複數Chebyshev 官能(Chebyshev sense )中便能獲得最佳的等漣漪濾波器。換句話說,如果一個適當的遞迴過程予以應用在此,則在複數Chebyshev 官能中便可獲得等漣漪濾波器。 

英文摘要: In this project, we present a new nonlinear fuzzy filter with genetic algorithm (GA) approach for image processing in a mixed noise environment, where both additive gaussian noise and non-additive impulsive noise may be present. We make use of the local statistics to retain the membership function of a fuzzy filter for image processing to remove both Gaussian noise and impulsive noise while preserving edges. The real GA approach is extended to complex cases for designing arbitrary complex finite-duration impulse response (FIR) digital filters. By minimizing a quadratic measure of the error in the passband and stopband, the chromosomes are evolved to get the complex filter coefficients. The method is based on an iterative complex GA approach to obtain a least squares approximation to the desired complex frequency responses and the weighting function at each iteration is updated using the result of the previous iteration, which leads to the optimal approximation in the complex Chebyshev sense. Namely, if an appropriate iterative process is used, equiripple filters in the complex Chebyshev sense can be obtained. 

 

 

科資中心編號: RB9102-0046 

題名: 智慧型影像處理於橋樑維護與檢驗技術之研究

Analysis of the Intelligent System of Image Processing to Maintain and Inspect Techniques of Bridges 

研究者: 黃世昌 ; 張陸滿

Huang, Shyh-Chang ; Chang, Luh-Maan 

機構名稱: 交通大學土木工程系 

經費來源: 行政院國家科學委員會 

補助金額: 新台幣 362 仟元 

補助編號: NSC89-2211-E009-068 

研究日期: 2000/08 - 2001/07 

出版日期: 2001/07 

頁冊數: 6頁 

原件提供: 行政院國家科學委員會科學技術資料中心, NSC89-2211-E009-068 

科資分類號: IG0003 

關鍵詞: 影像處理 ;橋梁檢測 ;橋梁管理 ;類神經網路

Image Processing ;Bridge Inspection ;Bridge Management ;Artificial Neural Network 

中文摘要: 近年來,由於電腦技術的發展讓影像處理技術具有實用性,而不需花費昂貴成本,也因為電腦處理器運算能力的不斷提昇,讓我們可以利用影像處理裝置與相關影像處理技術與軟體,將複雜的影像加以分析。而本研究在探討一種智慧型判斷模式,其結合影像處理與類神經網路,對於鋼構橋樑表面塗裝鏽蝕狀況進行辨認與量測,在系統中利用一些案例來訓練類神經網路,使其具有人類的經驗,故此系統與現行商業影像處理系統的主要差異在於其運用類神經網路,具有自我學習能力,並且具有容錯性。在本文中首先對於鋼構橋樑塗裝系統與鏽蝕機制加以介紹,並對於研究中相關運用之影像處理技術與方法予以說明,再者探討類神經網路各個網路架構與演算法,在研究中利用案例學習以建構網路架構並進行案例測試,最後就測試結果加以探討。 

英文摘要: Intelligent computerized system can simulate human expertise as well as analyze and process vast amounts of data instantaneously. This report presents a hybrid intelligent computerized system for bridge surface quality assessment. This system can be assessed to identify and measure the steel bridge coating conditions and defects through computers to analyze image of the areas. Moreover, neural network are used to train the system to automate the image processing and replicate the experts' knowledge in identifying the defects. The major difference between the proposed system and the existing commercial image processors is that the model has the intelligent ability to self-learn through neural networks and makes the decision of accepting or rejecting the assessed quality with pre- known risks. Finally those cases are successful to apply image processing and neural network techniques for bridges surface quality assessment to make the process objective, quantitative, consistent, and reliable. 

 

 

[科資中心編號]RB8910-0286

[題    名]

線性四元樹儲存類似影像之研究

A Study on the Linear Quadtree Representation for Storing Similar

Images

[研 究 者 ]

林聰武

Lin, Tsong-Wuu

[機構名稱  ]東吳大學資訊科學系(SCUCCIS)

[經費來源  ]行政院國家科學委員會(NSC)

[補助金額  ]376000

[補助編號  ]NSC88-2213-E031-002

[研究開始日期]1998/08

[研究結束日期]1999/07

[語    文]中文

[建檔日期  ]2001/06

[出版情形  ]完成日期:民國88年12月

[原件提供  ]行政院國家科學委員會科學技術資料中心

NSC88-2213-E031-002

[科資分類號 ]IA0403

[關 鍵 詞 ]

線性四元樹; 相似影像; 影像壓縮; 影像處理; 覆疊

 

Linear quadtree; Similar image; Image compression; Image processing;

Overlapping

 

[摘    要]

      連續影像擁有一些並未存在於個別影像的額外資訊,使得連續影像形成

一個逐漸改變的資料集,所有的動態改變資訊都完整地儲存於該資料集中。一個逐漸改變的資料集往往佔用相當大的記憶體空間,除非它將相同的部份與予覆疊。植基於線性四元樹及覆疊的概念,一個新的表示法被提出來儲存一串列的相同黑白影像,比較這個方法與其它已存在覆疊方法的實驗已經完成,由該實驗,證實這個新方法較其它現有方法好。

 

      Consecutive images keep some extra information which is not

stored in each of them. These consecutive images form a gradually

changing data set.The dynamic changes are kept wholly in such a

data set. A gradually changing data set occupies a large amount of

memory space unless it overlays the same parts. Based on the linear

quadtree structures and overlapping concepts, a new representation

is proposed for storing a sequence of similar binary images.

Experiments have been made to compare our representation with other

overlapping structures. From the experiments, our representation is

shown to be better than other representations.

 

科資中心編號: RB8907-0071 

題名: 影像密碼系統之快速加密法研究

A Fast Encryption Algorithm for Image Cryptosystems 

研究者: 黃明祥

Hwang, Min-Shiang 

機構名稱: 朝陽科技大學資訊管理技術系(CYUTINA) 

經費來源: 行政院國家科學委員會(NSC) 

補助金額: 新台幣 900 仟元 

補助編號: NSC86-2621-E324-001-T 

研究日期: 1997/02 - 1997/07 

頁冊數: 29頁 

原件提供: 行政院國家科學委員會科學技術資料中心 NSC86-2621-E324-001-T 

科資分類號: IB0002;IA0403 

關鍵詞: 密碼學 ;影像壓縮 ;影像處理 ;資料安全 ;影像密碼 ;加密

Cryptography ;Image Compression ;Image Processing ;Data Security ;Image Cryptosystem ;Encryption 

中文摘要: 早期從事密碼研究工作者大多集中在文字資料上,事實上,在網際網路除了傳輸文字(Text)資料外,尚有最吸引人之圖文並茂影像(Image)資料。當此影像資料為一重要機密資料,如軍方佈置圖、銀行建構圖、衛星拍攝圖等等,如何有效在網路上安全傳送此類資料,將是本文的重點。研究影像資料密碼系統,可從下列三方面進行:1.將影像資料視同文字資料:亦即,將二維圖像資料轉成一維串列資料,再將此資料視同文字資料做Stream或Block加解密處理。此密碼系統由於影像圖像通常很大(n*n pixel),因此,加解密將花費很多時間做計算。2.利用影像特性,先將圖像資料做影像壓縮(Image compression),再將此壓縮過影像資料,以文字資料方式做加解密。這種方式雖然比前述1方法要節省很多加解密時間,但由於壓縮後資料比起文字資料還是很大,因此,直接將壓縮過資料作加密,依然花費很多時間做運算。3.利用影像特性先將圖像資料做影像壓縮,再將壓縮用Table資料以文字資料方式做加解密。這種方式比前二種方法均節省許多加密時間,原因是壓縮用Table通常遠小於圖像資料。評估上述各種方法後,本文採用向量量化法高效率壓縮法、存取控制矩陣及密碼學理論方法。向量量化法之基本原理是利用一所謂號碼簿(Codebook)以取代具有相似特徵的訊息。加解密資料可將號碼簿之Index以傳統密碼系統處理,亦將此號碼簿本身做處理,但由於codebook之index為影像圖檔之指標序號,此序號之size依原圖而定,一般而言,比codebook之size大很多,因此,若直接從codebook做加解密,其效率較高。 

英文摘要: Implementations of cryptosystems for text data and image data are different. There are two major differences. One difference is that the size of image data is usually much larger than that of text data. The other is that plain data rarely permits loss when a compression technique is used, but image data does. In this paper, we propose a fast encryption algorithm for image cryptosystems. Our method is based on vector quantization, which is one of the popular image compression techniques. Our method can achieve the following two goals. One goal is to design a high security image cryptosystem. The other goal is to reduce computational complexity of the encryption and decryption algorithms. 

 

 

[科資中心編號]RB8903-0049

[題    名]

保持線條清晰的影像放大方法

Sharpness Preserving Image Enlargement

[研 究 者 ]

呂嘉穀

Leu, Jia-Guu

[機構名稱  ]中興大學統計系(NCHLSTA)

[經費來源  ]行政院國家科學委員會(NSC)

[補助金額  ]156000

[補助編號  ]NSC88-2213-E005-A-001

[研究開始日期]1998/08

[研究結束日期]1999/07

[頁 冊 數 ]4頁

[語    文]英文

[建檔日期  ]2000/10

[出版情形  ]完成日期:民國88年12月

[原件提供  ]行政院國家科學委員會科學技術資料中心

NSC88-2213-E005-A-001

[科資分類號 ]IA0403

[關 鍵 詞 ]

影像放大; 影像處理; 邊線偵測

 

Image enlargement; Image processing; Edge detection

 

[摘    要]

      在多媒體文件製作過程中,我們有時需要將一個影像放大以配合版面的

設計。傳統像素重複法所產生的線條會呈現鋸齒狀,並不平滑。而以內插法為基礎的方法所產生的放大影像中的線條能夠既平滑又清晰。

      在本法中我們首先找出所要放大影像中所有的邊像素,然後我們為每一

個邊像素導出五個坡面參數。這五個參數值就用來描述該像素所歸屬之邊線段以及該像素在邊上的位置。其後我們就用原影像中邊像素之參數來為放大影像中對應位置上的每個邊像素也導出一組參數值。放大影像中的邊就可以利用這些參數值來產生。如果我們在放大影像時保持各邊線之原有邊寬值,則放大影像中的邊線將會與原影像中的邊線一般清晰。

 

      In multimedia applications sometimes we need to enlarge an

image for it to fit the overall page layout. The traditional pixel

repetition method produces edges that are jagged, and the

interpolation-based methods tend to produce blurry edges. In this

project, we present an approach that is able to maintain both the

continuity and the sharpness of the edges when enlarging an image.

      In the approach we first locate the edge pixels in an image.

We then extract for each edge pixel five ramp parameter values.

These five values together characterize locally the edge segment

associated with the pixel and the location of the pixel on the edge.

Next, we use the parameter values of the edge pixels in the

original image to derive parameter values for the edge pixels in

the enlarged image. The intensity of the edge pixels in the

enlarged image can then be determined from their parameter values.

If we keep the ramp width parameter values at the same level for

the edge pixels in the enlarged image, the enlarged edges will be

as sharp as the edges in the original. The experiments show that

the suggested ramp model based approach indeed produces

enlargements with continuous and well defined edges.

 

[科資中心編號]RB8805-0116

[題    名]

分散式立體影像重建系統(II)

Distributed 3D Image Reconstruction System (II)

[研 究 者 ]

陳立祥; 陳志鴻

Chen, Lih-Shyang; Chen, Jyh-Hong

[機構名稱  ]成功大學(NCKU)

[經費來源  ]行政院國家科學委員會(NSC)

[補助金額  ]5681000

[補助編號  ]NSC87-2622-E006-007

[研究開始日期]1997/10

[研究結束日期]1998/09

[頁 冊 數 ]47頁

[語    文]英文

[建檔日期  ]1999/11

[出版情形  ]完成日期:民國87年12月

[原件提供  ]行政院國家科學委員會科學技術資料中心

NSC87-2622-E006-007

[科資分類號 ]IA0403

[關 鍵 詞 ]

影像重建; 超錄製; 影像處理; 三維影像; 超控制

 

Image reconstruction; Hyper record; Image processing; Three

dimensional image; Hyper control

 

[摘    要]

      超控制(hyper-control)的基本理論是利用一份稱之為超控制文件

(hyper-control document)的多媒體文件去控制其他的應用程式。在本年度的

計畫中,我們提出一個新的理念,稱為超錄製(hyper-record),以方便超控制文

件的建立。我們主要是利用命令處理樣式(Command Processor Pattern),並配

合其他相關連樣式,來發展我們的醫學影像系統Discover;它可以同時支援超控制和超錄製的機制。在所給的例子中,我們會舉例描述Discover和超控制文件之間的互動關係。而對於我們使用樣式技術的經驗,也會在此報告中敘述。

 

      The basic idea of hyper-control is to use a multimedia

document, called a hyper-control document, to control other

application systems. In this project, we propose a new concept,

called hyper-record, to facilitate the creation of hyper-control

documents. Based on an extension of the Command Processor pattern

and other related patterns, we have developed a medical image

system, called Discover, which can support both the hyper-control

and hyper-record mechanisms. Examples are given to illustrate the

mutual relationship between Discover and its hyper-control

documents. Our experience in using patterns is also addressed.

 

[科資中心編號]RB8805-0138

[題    名]

碎形影像壓縮之快速演算法則

Fractal Image Compression: A Fast Algorithm

[研 究 者 ]

鄭志宏; 張肇健

Jeng, Jyh-Horng; Chang, Trieu-Kien

[機構名稱  ]義守大學資訊工程系(ISUIFE)

[經費來源  ]行政院國家科學委員會(NSC)

[補助金額  ]265000

[補助編號  ]NSC87-2213-E214-007

[研究開始日期]1997/08

[研究結束日期]1998/07

[頁 冊 數 ]10頁

[語    文]中文

[建檔日期  ]1999/11

[出版情形  ]完成日期:民國88年2月

[原件提供  ]行政院國家科學委員會科學技術資料中心

NSC87-2213-E214-007

[科資分類號 ]IA0403

[關 鍵 詞 ]

碎形影像壓縮; 演算法; 離散餘弦轉換; 影像處理

 

Fractal image compression; Algorithm; Discrete cosine transform

(DCT); Image processing

 

[摘    要]

      碎形影像壓縮技術有著極高的壓縮率,而且解壓縮後能保有相當好的畫

,但壓縮時間極長為其缺點,其主要原因為相似子影像之尋找。在本計畫使用一套快速演算法,這套快速演算法的主要目的在於降低碎形影像壓縮在尋找相似子影像的區塊搜尋時所花的運算時間。

      由於在基本的做法上,碎形影像壓縮的編碼法,主要是利用影像間的自我

相似性來達成。而為了要找出最具自我相似性的子影像,子影像的0.degree.、90.degree.、180.degree.、270.degree.旋轉以及其對直線Y=X鏡射,所產生出

的八個方向的子影像區塊將被使用。但是由於分析子影像的自我相似性需花費龐大的運算時間,因此如何加速運算時間就是本計畫研究的主要目標。

      在本計畫中,DCT將被使用。所以自我相似性的影像分析將經由DCT,由時域轉換到頻域上。藉由分析上述八個方向的頻域資料中可以發現,這八個方向的頻域資料有著正負號變化的重複性,故多餘的重複運算將可被減少。由於人類的視覺系統對高頻影像變化較不敏感,故本計畫也對只使用部分頻段資料為搜尋依據時,其運算時間及失真率的影響加以分析。由軟體模擬的結果顯示,在相同的PSNR、壓縮比以及無使用分類法的情況下,這新的演算法在尋找相似子影像的區塊搜尋上所花的運算時間比基本的作法減少了至少三倍之多。若只使用部分頻段資料為搜尋依據,其所需的運算時間更可明顯降低。

 

      Fractal Image Compression provides high compression ratio and

high restore qualities. Yet, it is time consuming in the

compression process. This is because of the search of the 'similar'

subimages. In this project, a fast algorithm has been developed.

The main goal of this algorithm is to reduces the computation time

on the search of subimages, which has the most self-similarity.

      In the baseline system, the encoding method of the Fractal

Image Compression is done by searching the self-similarity. For

finding the subimages that has the most similarity, the eight

orientation of subimage are used, which is generated by rotating

the blocks counter clockwise at angles 0.degree., 90.degree.,

180.degree. and 270.degree. and flipping with respect to the line

Y=X. But it will spend a mount of computation time for the

similarity analysis. So, the major goal of this project is to speed

up the computation time on encoding process.

      In this project, the DCT is used. Though the DCT, the

similarity analysis of subimage will be transformed form the time

domain to the frequency domain. According to the analysis of the

frequency data form eight orientations, the redundancy of the

frequency data are discovered. So, the redundant computations are

eliminated. Because that, the human visual system is non-sensitive

by the higher frequency image. So, we will also use part of the

frequency data to improve the encoding time while the PSNR is not

changed. A software simulation is presented in this project which

shows that, with the same PSNR, the fast encoding method is at

least 3 times faster than that using the base line algorithm. If we

only use a part of the frequency data, the computation time need

for search cloud be reduced more.

 

[科資中心編號]RB8805-0005

[題    名]

利用禁忌搜尋求解聚類問題

Using Tabu Search Techniques to Solve the Clustering Problems

[研 究 者 ]

林聰武

Lin, Tsong-Wuu

[機構名稱  ]東吳大學資訊科學系(SCUCCIS)

[經費來源  ]行政院國家科學委員會(NSC)

[補助金額  ]259700

[補助編號  ]NSC87-2213-E031-002

[研究開始日期]1997/08

[研究結束日期]1998/07

[頁 冊 數 ]4頁

[語    文]英文

[建檔日期  ]1999/11

[出版情形  ]完成日期:民國87年10月

[原件提供  ]行政院國家科學委員會科學技術資料中心

NSC87-2213-E031-002

[科資分類號 ]IA0101; IA0406; IA0403

[關 鍵 詞 ]

群聚分析; 塔布搜尋法; 圖形辨識; 影像處理

 

Cluster analysis; Tabu search; Pattern recognition; Image processing

 

[摘    要]

      群聚運算將m 個物件分配到g個群中,同一個群中的成員具有類似的特

,將物件的特徵以n度空間的點座標來表示,因此點與該群中心的距離可以表達物件在該群的適當性,群聚運算的目標在於將所有點至其群中心距離的總和最小化。這個問題有許多的局部最小值,一個植基於禁忌搜尋的增強演算法被用來解決該問題,一些產生嘗試解的策略被提出來以增進其效力。由實驗結果得知我們的方法較現存的方法有效且花費較少的執行時間。

 

      The clustering operation is to classify m objects into g

groups whose members are similar in the interesting features. The

features of an object are represented as a point within an

n-dimensional Euclidean space. The distance between a point and the

center of its group indicates fitness of the object within the

group. The objective is to divide these objects into groups such

that the total sum of these distances is minimized. There are many

local minima in the problem. An improved algorithm is proposed to

solve the problem based on the tabu search technique. Some

strategies for generating trial solutions are presented to improve

its effectiveness. From the experiments, our strategies are very

powerful than existing methods.

 

[科資中心編號]RB8803-0126

[題    名]

隨機通道下的影像復原技術

Restoration of Images Distorted by Random Channels

[研 究 者 ]

洪賢昇

Hung, Hsien-Sen

[機構名稱  ]台灣海洋大學電機工程系(NTOSELE)

[經費來源  ]行政院國家科學委員會(NSC)

[補助金額  ]182300

[補助編號  ]NSC87-2213-E019-003

[研究開始日期]1997/08

[研究結束日期]1998/07

[頁 冊 數 ]5頁

[語    文]中文

[建檔日期  ]1999/09

[出版情形  ]完成日期:民國87年10月

[原件提供  ]行政院國家科學委員會科學技術資料中心

NSC87-2213-E019-003

[科資分類號 ]IA0403; IE0501

[關 鍵 詞 ]

影像處理; 影像復原; 隨機通道; 完最小平方演算法

 

Image processing; Image restoration; Random channel; Total least

square algorithm (TLS algorithm)

 

[摘    要]

      由於影像在隨機通道下傳送時容易造成雜訊干擾和模糊效應,本計畫採

用本人提出的限制性全體最小方差法(CTLS)[1],做為影像復原的理論基礎。然而CTLS演算法在二維影像復原時所需的計算量頗大,為了降低計算複雜度,本計畫提出共軛梯度和離散傅立業轉換域兩種實現CTLS的方法。

      此外,為了消除共軛梯度法於影像邊緣所產生的環狀假象,採用權重空間

範數的觀念,使得演算法可依影像的相鄰像素變化的情形,來調整調制項的權重值,而達到消除環狀假象的目的。最後以電腦模擬的結果,來驗證所提演算法的效能。

 

      The method of Constrained Total Least Squares (CTLS) proposed

by the author [1] is adopted in this project as a framework for

restoring images degraded by noise and blur, when transmitted over

a random channel. However, CTLS requires large computational

complexity for the purpose of 2-D image restoration. In order to

reduce computational complexity, the conjugate gradient method and

the DFT frequency-domain method are proposed for the realization of

CTLS in the project.

      In addition, the concept of matrix norm in the weighting

space is adopted in the regularization term of the conjugate

gradient method in order to eliminate ringing artifacts near the

boundary (edge) regions. This mechanism provides the underlying

algorithm with adaptation feature; the weights in the

regularization term can be adjusted according to the local property

of pixel value changes in the image of interest. Finally, computer

simulation results are provided to show the effectiveness of the

proposed algorithms

 

[科資中心編號]RB8803-0105

[題    名]

影像曲面之分析、表示及處理方法之研究

The Analysis, Representation and Manipulation of Image Surfaces

[研 究 者 ]

王聖智

Wang, Sheng-Jyh

[機構名稱  ]交通大學電子工程系(NCTEELE)

[經費來源  ]行政院國家科學委員會(NSC)

[補助金額  ]1146900

[補助編號  ]NSC87-2213-E009-066

[研究開始日期]1997/08

[研究結束日期]1998/07

[頁 冊 數 ]4頁

[語    文]中文

[建檔日期  ]1999/09

[出版情形  ]完成日期:民國87年11月

[原件提供  ]行政院國家科學委員會科學技術資料中心

NSC87-2213-E009-066

[科資分類號 ]IA0403; IE0501

[關 鍵 詞 ]

高曲率特徵點; 影像表示法; 影像特徵擷取; 影像處理; 影像分割; 影像表面

 

High curvature feature point; Image representation; Image feature

extraction; Image processing; Image segmentation; Image surface

 

[摘    要]

      在這個研究中,我們提出一個以影像高曲率特徵點為基礎的影像表示法

。利用這些特徵點,影像的特性可以很正確地被此表示法所掌握,並進一步使得我們在影像處理的操作上更加方便。對於特徵點萃取的方式,我們也做了定性與定量的探討。最後,我們以一個將影像用所提出的高曲率特徵點表示,並在其後將之還原的完整流程,作為在實際的應用上的展示。

 

      In this project, we present a new image representation method

based on high-curvature features in an image. By using these

features, an image can be precisely characterized by the proposed

representation scheme. Moreover, we are capable of manipulating the

image in an easier way. The extraction of image feature points is

also discussed both qualitatively and quantitatively. At last, a

complete flow to represent an image in the proposed approach first

and then to reconstruct the image back is performed as an example

for practical applications.

 

[科資中心編號]RA8604-2088

[題    名]

立體影像重建之技術研究

3D Image Reconstruction Techniques.

[研 究 者 ]

陳立祥

CHEN,LIH-SHYANG

[機構名稱  ]成功大學電機工程系(NCKEELE)

[經費來源  ]行政院國家科學委員會(NSC)

[補助金額  ]280000

[補助編號  ]NSC83-0404-E006-059

[研究開始日期]1994/02

[研究結束日期]1995/01

[頁 冊 數 ]26頁

[語    文]中文

[建檔日期  ]1997/11

[出版情形  ]完成日期:民國84年7月

[原件提供  ]行政院國家科學委員會科學技術資料中心

NSC83-0404-E006-059

[科資分類號 ]IA0403;EA0204

[關 鍵 詞 ]

影像重建;超音波影像;數位影像處理;三維物體

 

Image reconstruction;Ultrasound image;Digital image processing;3D

object

 

[摘    要]

      目前醫學界使用最普遍、最便宜、最快速、最安全的非破壞性診斷內臟儀器就是超音波掃描。但其掃出的影像卻是2D剖面圖,若能利用電腦將其轉成3D圖形,將能大幅提高醫師的診斷力和超音波掃描的價值。超音波掃描與MRI(Magnetic resonance imaging)、CT (Computed tomography)之不同在於:前者適用於靜態、動態內臟掃描,後者僅能作靜態掃描。比如超音波能做心臟透視,MRI則否。且其速度快能爭取時效性,是超音波掃描的優點。但其影像品質卻相當差,且其掃描方式與MRI不同,使得其重建成3D圖形過程困難重重。

      所以,本計畫以左心室超音波掃描為例,來闡明如何將2D超音波掃描圖形,重建成3D物件。超音波掃描重建3D步驟如下所示:(1)將超音波掃描機器產生的資料轉成我們需要的資料。(2)將左心室的輪廓找出來。要做到全自動找出左心室輪廓,相當困難。而本計畫先以一些數位影像處理的技巧改善影像品質,再以本計畫所提出方式找出左心室的輪廓。(3)將所有左心室輪廓根據其掃描角度予以旋轉建立3D物件。事先建好一些旋轉計算所需資料的表,以減少3D重建的運算量,再予以旋轉重建。 雖然本計畫以左心室為例,但並不以此為限,舉凡超音波掃描資料皆可使用此3D重建之演算法,快速地轉成3D圖形。

 

      The ultrasound sector scanner is the most pervasive, cheapest,

fastest, and safest tool for medical diagnosis of a human body

compared with other imaging modalities. Since the ultrasound images

are basically the 2D cross-sectional images, the accuracy and

efficiency of physicians' diagnoses can be significantly improved if

the 3D objects can be reconstructed from those 2D images by computers.

The ultrasound images are different from MRI (Magnetic Resonance

Imaging) and CT (Computed Tomography) images in the sense that they

are real-time images. Therefore, they can be used effectively for

imaging moving objects, such as a beating heart. In spite of the

merits of the ultrasound images, their 3D image reconstruction is

much more difficult compared with those of the MRI and CT images due

to the poor image quality and resolution of the 2D ultrasound images.

      In this project, the algorithms for the 3D image reconstruction

of the ultrasound images of left-ventricular are proposed. The major

algorithm can be divided into three steps. (1) Transforming the

original data of ultrasound sector scanner into certain format for

further processing. (2) Extracting the contour of left-ventricle

automatically. We improve quality of images by some digital image

processing techniques. Then, we propose an efficient algorithm to

extract automatically the contour of left-ventricle. (3)

Reconstructing the 3D object using all the contours of the left-

ventricle images taken at the same time instant of different periods

with different orientations. We propose a table look-up algorithm to

reconstruct the 3D object from the 2D contours. The table look-up

method can reduce a lot of computation time. After the 3D object is

reconstructed, it can be displayed on the screen using 3D computer

graphics techniques. The volume of the 3D object can be calculated at

the reconstruction time.

      The 3D reconstruction algorithm for the 2D ultrasound images

can be applied to any other ultrasound images to reconstruct the 3D

objects.

 

[科資中心編號]RA8507-2090

[題    名]

以小波轉換作影像處理

Image Processing Using the Wavelet Transform.

[研 究 者 ]

賴坤財

LAY,KUEN-TSAIR

[機構名稱  ]台灣工業技術學院電子工程技術系

(NTITENT)

[經費來源  ]行政院國家科學委員會(NSC)

[補助金額  ]172000

[補助編號  ]NSC83-0404-E011-037

[研究開始日期]1994/02

[研究結束日期]1995/07

[出版日期  ]1995/08

[頁 冊 數 ]164頁

[語    文]中文

[建檔日期  ]1997/01

[出版情形  ]研究報告,164頁,民國84年8月

Report, 164p., Aug. 1995

[原件提供  ]行政院國家科學委員會科學技術資料中心

NSC84-2213-E011-059

[科資分類號 ]IA0403

[關 鍵 詞 ]

波元轉換;影像處理;影像壓縮;多解析度;局部

 

Wavelet transform;Image processing;Image compression;Multiresolution;

Localization

 

[摘    要]

      影像壓縮是降低影像之傳輸頻寬和儲存空

間最基本的技術。近年來由於小波理論的成熟

發展,已經成功的應用在信號處理各種領域上,

同時在影像壓縮技術上的應用也已受到廣泛的

注意。有鑑於此,在本計畫中,我們提出三種基

於小波分析的影像編碼架構,分別是基於小波

轉換、小波封包以及小波全分解等,並結合位

元率---失真理論以達到編碼之最佳化。實驗的

結果顯示,我們提出的編碼架構,不管在主客觀

評估上都可以獲得令人滿意的結果。此外,我

們也探討了小波分析的幾個重要的特性,諸如

垂交性、對稱性以及平滑度等對編碼效率實際

影響的程度。

 

      Image compression is essential in reducing the amount of the

storage space and the bandwidth of transmission of digital images.

      The application of the wavelet theory to digital compression

has become a hot topic in recent years. Accordingly, based on the

wavelet analysis, three encoding schemes are proposed to compress

images. Experimental results show that the proposed coding schemes

produce satisfactory reconstructed images, as measured both

mathematically (in terms of signal-to-noise ratio and compression

ratio) and visually. The relationships between coding efficiency and

relevant properties of wavelets (orthogonality, symmetry and

regularity) are also investigated.

 

[科資中心編號]RA8604-2352

[題    名]

以鬆弛法做微波影像分割

Microwave Image Segmentation by Relaxation.

[研 究 者 ]

曹建和

TSAO,JEN-HO

[機構名稱  ]台灣大學電機工程研究所(NTUGELEG)

[經費來源  ]行政院國家科學委員會(NSC)

[補助金額  ]233000

[補助編號  ]NSC83-0404-E002-036

[研究開始日期]1993/08

[研究結束日期]1994/07

[頁 冊 數 ]76頁

[語    文]中文

[建檔日期  ]1997/11

[出版情形  ]完成日期:民國83年7月

[原件提供  ]行政院國家科學委員會科學技術資料中心

NSC83-0404-E002-036

[科資分類號 ]IE0509;IE0501;IA0403

[關 鍵 詞 ]

雷達影像;影像分割;鬆弛法;影像處理

 

Radar image;Image segmentation;Relaxation method;Image processing

 

[摘    要]

      用雷達波成像最大的困難在於其波長太長,所以其影像沒有光學影像的解析度,導致影像單元的灰度值(Gray level)會比光學影像更大的誤差,若想要直接由此而獲得我們要的特徵(Feature),是一件困難的事,所以必須藉由影像處理的技巧得到較好的影像,以作為後端處理之用。

      在本論文裡,主要是利用空間的資訊(Contextual information)去減少影

像單元的錯誤,其中最為大家所熟知的是鬆弛法(Relaxation)。鬆弛法在光學影像處理方面已有良好的成果,將之用於雷達影像的效果,將在本論文有詳盡的討論,而其他的一些方法亦將一併提出。當影像單元的錯誤減少後,接著利用邊緣尋找法或圍線尋找法,得到影像的邊緣或圍線,然後以此當做後端處理的特徵。在此期望對於雷達影像的處理有所貢獻。

 

      One of the difficulties in target identification based on its

microwave image is target boundary detection. Since the gray level of

microwave image is contaminated by sidelobe responses, to get proper

target boundary, its image must be segmented first. In this report,

the relaxation method is applied on high resolution microwave images

to improve the definition of target and background.

      The key issue in relaxation method is how to utilize the

contextual information of an image. For the special case of microwave

image, a workable algorithm deduced from several try and error

studies is given for extracting the contextual information. It is

concluded that the success of microwave image segmentation depends

greatly on the capability in finding a proper compatibility for the

relaxation procedure.

 

[科資中心編號]RA8501-2052

[題    名]

建立於四分樹的形態影像運算

Mathematical Morphology Based upon Quadtree.

[研 究 者 ]

周本生

CHOW,BEN-SHUNG

[機構名稱  ]中山大學電機工程系(NSYEELE)

[經費來源  ]行政院國家科學委員會(NSC)

[補助金額  ]212000

[補助編號  ]NSC82-0115-E110-143

[研究開始日期]1993/02

[研究結束日期]1994/01

[出版日期  ]1994/05

[頁 冊 數 ]62頁

[語    文]中文

[建檔日期  ]1996/10

[出版情形  ]研究報告,62頁,民國83年5月

Report, 62p., May 1994

[原件提供  ]行政院國家科學委員會科學技術資料中心

NSC82-0115-E110-143

[科資分類號 ]IA0403

[關 鍵 詞 ]

影像處理;四元樹;電腦視覺;影像壓縮

 

Image processing;Quadtree;Computer vision;Image compression

 

[摘    要]

      數理形態處理法(Mathematical morphological

processing),近來在工業檢測與醫學影像上的運用

有很大的發展,其基本運算為擴張(Dilation)、侵

(Erosion)、開放(Opening)和閉鎖(Closing),而這些

運算完全建立於集合的交集、聯集、補集上,

對黑白影像而言,因為可以用1與0來表示。所以

適合於布林邏輯運算,而計算機即是此種運算

的結構,自然適合此類運算。四分樹(quadtree)資

料結構,在影像壓縮(Image compression)與影像分類

(Image segmentation)上,已有成功的運用,由於四分

樹法可將同顏色之圖素合併成一最大方塊,同

時也可區別影像的圖形與背景。而擴張運算是

針對圖形部分作運算,侵蝕運算則是針對背景

部分作運算,應用四分樹架構便可使我們省下

可觀的運算次數。在形態運算上,方塊可以被

分解成水平和垂直線段的擴張運算,而在四分

樹架構只有幾類方塊,我們可以預先對這幾類

方塊做好擴張運算存放著,以便將來重複運算

,可以對映方式完成,這可幫助運算簡單化。

 

      Mathematical Morphology image processing has recently been

applied successfully to industry auto-inspection and medical image

processing. Morphology was originally a branch of biology that deals

with the forms and shapes of cells. The variation of forms for

objects is usually nonlinear, the computation cost of nonlinear

processing for conventional numerical image processing is very

expensive. However, morphology image processing treats the image

components as sets and deals with the change of shape very

efficiently. The basic image operations for morphology processing

are erosion, delation, closing and opening, which are totally built

upon the Boolean operation: set union, intersection and

complementation. This accounts for the efficiency of morphology

processing since computation for Von Neumann computer is based upon

logic operation.

      The quad-tree data structure has been successfully applied to

the fields of computer vision such as image segmentation and

compression. The quad-tree with its hierarchical data structures are

advantageous due to its ability to focus on the interesting subset

of the data. This focusing results in an efficient representation.

Thus, the quad-tree data. This focusing results in an efficient

representation. Thus, the quad-tree data structure are particularly

convenient for set operations. Therefore, the computation for

morphology processing will be facilitated by using the quad-tree

structure since the set operations are the basics of morphology

processing. In a summary, the purpose of this project are not only

to speed up the morphology processing but also to fulfil the needs

of data compression, which are usually contradict with each other.

 

[科資中心編號]RA8410-2085

[題    名]

碎形幾何在影像處理上的應用

Applications of Fractal Geometry in Image Processing.

[研 究 者 ]

薛元澤

HSUEH,YUANG-CHEH

[機構名稱  ]交通大學資訊科學系(NCTSIFS)

[經費來源  ]行政院國家科學委員會(NSC)

[補助金額  ]377000

[補助編號  ]NSC82-0408-E009-430

[研究開始日期]1993/02

[研究結束日期]1994/01

[出版日期  ]1993/01

[頁 冊 數 ]57頁

[語    文]英文

[建檔日期  ]1996/06

[出版情形  ]研究報告,57頁,民國82年1月

Report, 57p., Jan. 1993

[科資分類號 ]IA0403

[關 鍵 詞 ]

碎形幾何;影像處理;影像分析;影像壓縮

 

Fractal geometry;Image processing;Image analysis;Image compression

 

[摘    要]

      近年來,碎形幾何已經被廣泛地應用在科

學的各領域中。在這份報告裡,我們嘗試去研

究碎形的特性以及它在影像處理上的應用,兩

個利用碎形特性來處理的主題,影像分析及影

像壓縮,將被詳細的探討。在影像分析的問題

,我們提出兩種方法來做紋理影像的分析和

分類,其中一種是利用多個空間填滿線條來計

算影像的長度,另一種則是用不同解析度下影

像所呈現的不同面積來做紋理影像分析的依據

。實驗結果獲得很高的正確分類辨識率。在影

像壓縮的問題上,我們提出一種碎形內插法來

達到影像壓縮的目地,這種方法是利用影像區

域間的自我相似性來預估內插影像點的值。利

用這種方法所得到的內插影像能保持影像原有

的邊而不會被模糊掉。另一個好處是這個方法

容易實作且執行速度快。在這份報告裡,我們

同時比較這個方法以及線性內插法所得到的結

果。

 

      Recently, fractal geometry has been widely used in different

areas of science. In this report, we have attempted to study fractal

properties and their applications to image processing. Two major

issues, image analysis and image compression, based on fractal

geometry are discussed. In image analysis, we present two methods,

multiple space-filling curves and multiresolution approaches, to

analyze and classify texture images. Experimental results reveal

high recognition rates by using the proposed methods. In image

compression, we have introduced the basic concepts of generally used

fractal compression methods. In addition, an interpolation method

based on self-similar property of images is proposed to achieve the

compression purpose. The resulting images using the proposed method

show more enhanced and better quality than using linear

interpolation.

 

[科資中心編號]RA8401-2002

[題    名]

碎形幾何在影像處理及識別上之研究

A Study of Fractal Geometry on Image Processing and Recognition.

[研 究 者 ]

王玲玲

WANG,LING-LING

[機構名稱  ]清華大學資訊科學研究所(NTHUCOS)

[經費來源  ]行政院國家科學委員會(NSC)

[補助金額  ]418000

[補助編號  ]NSC82-0408-E007-304

[研究開始日期]1993/02

[研究結束日期]1994/01

[出版日期  ]1994/01

[頁 冊 數 ]43

[語    文]英文

[建檔日期  ]1995/10

[科資分類號 ]IA;IA0403

[關 鍵 詞 ]

碎形影像壓縮;疊代函數;影像處理;模式識別

 

Fractal image compression;Iterated function;Image processing; Pattern

recognition

 

[摘    要]

      碎形壓縮在目前是一個新穎的影像壓縮技 術,並在實際

的影像壓縮領域應用上展現其嶄 新而強大的能力。此種影

像壓縮技術可以產生 高達到一萬比一的影像壓縮率。在此

技術中, 影像資料並非由一系列之影像點(Pixel)矩陣所 組成,

而是由一個相當緊密的數值集合所構成 的。這個數值集合

即為反複函數系統數碼 (Iterated function system code),而且這些數

值會被用 來建構復原影像的各個區塊,再逐一建構成完 整的

復原影像。

      本研究提出一種將碎形幾何應用在影像處 理及識別上

的新研究,使碎形幾何的應用範圍 更加廣泛。我們在進行此

研究時,是先利用碎 形幾何來作影像的壓縮工作,當此工作完

成之 後,即可利用影像經壓縮後所產生的反複函數 系統數碼

作後續的應用工作,而這些數碼則是 作影像復元(Image

reconstruction)所需的基本資料, 透過改變這些數碼的值,我們可

以在復元影像 上得到一些特定的影像處理效果,這即是本研

究的主題。

      這些影像處理工作包括對原始影像上雜訊 的去除(Noise

removal)、影像上邊界的偵尋(Edge detection)、邊界的放大或縮小

(Edge magnification/Demagnification)以及對影像上不同紋理 的切割

(Texture segmentation)等。實驗結果將會在 文章後列出,以顯示本

研究所提出的方法確實 有其實用性。

 

      The fractal image compression scheme, a new kind of compression

technique, has revealed its brand new and powerful capability in the image

compression category. The method has yielded compression ratio in excess of

10,000 to 1. An image is not represented in an array of pixels; relatively

compact set of numbers, called iterated function system (IFS) codes, are used

to code the segments that compose the image.

      In this paper, emphasis is not put on the fractal image compression

technique; instead, several characteristics of the IFS codes are proposed.

When decompressing an image using the IFS codes, some image processing tasks

can be done simultaneously. These tasks include noise removal, edge detection,

edge magnification or demagnification, and texture classification.

Experimental results are presented to show the feasibility of the proposed

approach.

 

 

[科資中心編號]RA8510-2093

[題    名]

動態影像區分及其應用

Motion Segmentation and Its Application.

[研 究 者 ]

杭學鳴

HANG,HSUEH-MING

[機構名稱  ]交通大學電子工程研究所(NCTUELEG)

[經費來源  ]行政院國家科學委員會(NSC)

[補助金額  ]481000

[補助編號  ]NSC83-0408-E009-012

[研究開始日期]1993/08

[研究結束日期]1994/07

[出版日期  ]1994/12

[頁 冊 數 ]75頁

[語    文]英文

[建檔日期  ]1997/05

[出版情形  ]研究報告,75頁,民國83年12月

Report, 75p., Dec. 1994

[原件提供  ]行政院國家科學委員會科學技術資料中心

NSC83-0408-E009-012

[科資分類號 ]IA0403

[關 鍵 詞 ]

影像處理;活動影像;動態影像區分;位移估測;

全域運動

 

Image processing;Moving image;Motion segmentation;Motion estimation;

Global motion

 

[摘    要]

      本報告的內容包含兩部分:第一部分是蒐

集、整理、比較現有之動態影像區分與物體位

移估測的演算方法;第二部分則是以取出全域

動態參數的方式來進行動態區分。

      1.運動訊息的取得對許多影像處理的應用

而言,是既基本且又重要的過程。欲得到正確

的運動訊息,物體位移估測與動態影像區分是

不可或缺的工具。然而兩者之間,存在有因果

循環的矛盾關係:良好的動態區分品質必須藉

助正確的位移估測;而缺少動態區分作為基礎,

又無法充分掌握物體運動的真實狀況。為了克

服上述的問題,實有必要對這兩種演算方法,做

深入的了解與探討。物體位移估測的技術發展

至今,已有二十五年的歷史。主要的方法可分

為三類:(1)區塊比對法---其特色為直觀、簡單,

因此廣受視訊編碼標準的青睞。(2)微分估測法

---此法建立在動態影像之時空關係的數學假設

,可進一步推導出遞迴演算法。(3)頻域估測

---有別於前兩種方法的出發點,在頻域中不

但可估測物體的運動向量,同時亦能提供探索

估測極限理論的管道。早期的動態影像區分演

算法多架構在相鄰畫面之靜態區分的結果上。

靜態區分的方法包括測量空間劃分、區域成長

、以及分裂合併術。在另一方面,由於計算機

速度與高精密度儲存技術的不斷演進,複雜的

動態分析已能實際應用在動態影像的處理上。

本報告中將對以下三類的動態分析法---形狀導

向、亮度導向、以及時空序列導向,評估其原

理及性能,作為設計或改進動態區分法的參考

      2.在這一部分的計畫中,我們根據全域運動

之資訊,提出一個新的動態影像區分方法。此

法由兩個步驟結合而成:第一是區塊比對式運

動估計,第二是全域運動歸原。其中每一張影

像可視為若干區塊的組合。我們考慮的全域運

動有三種,分別是推拉、旋轉搖動、以及水平

搖動。就整張圖片而言,推拉和旋轉搖動是唯

一且與物件無關;相反的,由水平搖動所造成的

運動向量卻和攝影機與物件之間的距離有關。

因此,物件與與水平搖動運動向量都可用來作

為動態影像區分的參數。其中第一步驟是利用

全搜尋式區塊比對來估測區域運動資訊,此步

驟所導出之運動向量包含推拉、搖動、以及物

件位移所複合而成的向量。第二步驟將區域運

動從全域運動中分離出來,而其分離的方法是

利用最小平方估測以及奇值分解的技巧來完成

。在此步驟中,我們推導出一數學模式來描述

此種包含推拉、搖動、以及物件位移的複合運

動向量,再根據物件指配法將具有相同全域運

動特性之區塊歸併為同一物件。實驗顯示出此

種運動歸原的方法能夠使由於推拉、搖動所造

成的運動向量熵值降低,其階可達到百分之三

、四十,因此可用以增加資料的壓縮效率。另

一方面,我們將此法運用到動態影像區分的方

法上,實驗結果發現,全域歸原的確可以將具有

相同全域特性的物件區分開來,故十分適合運

用來區分動態影像。

 

      This report contains two parts. Part 1 is a survey of various

motion estimation and segmentation methods. A new motion

segmentation algorithm based on global motion information extraction

is proposed in Part 2.

      1. Motion information extraction is fundamental and essential

for many image processing applications. It consists of two subjects:

motion segmentation and motion estimation. There exists a chicken

and egg dilemma between them: accurate motion information of every

pel is necessary for precise moving object segmentation, and precise

object boundaries are necessary for computing accurate motion

parameters. So far no good solution is provided for the dilemma. In

order to explore this problem further, we have studied and compared

the principles and performance of several well-known motion

segmentation and estimation algorithms. The goal of motion

estimation is to extract motion information of objects (pels) from a

time-varying image sequence. Motion estimation techniques can be

roughly classified into three groups: The block matching methods are

most intuitive and simplest. This is one of the key reason that it

has been widely used in video coding standards. The so-called

differential methods are derived based on an analytic assumption of

the spatial and temporal relationship of image sequences. From a

different point of view, the Fourier methods estimate motion vector

in frequency domain. It can provide insight on the theoretical

limits of motion estimation. Early motion segmentation algorithms

were proposed based on the image segmentation results between

successive picture frames. The spatial-domain image segmentation

methods includes measurement space clustering, region growing, and

split and merge methods. Because of the advance of high speed

computing and high density storage, motion analysis has become a

practical tool in image sequence processing. Three types of motion

segmentation schemes---feature-based, intensity-based, and image-

sequence-based approaches are described.

      2. A method of segmenting an image sequence into regions of

different moving objects based on extracting the global motion

information is proposed. It consists of two steps: (1) block-

matching motion estimation that estimates the displacement vectors

of image blocks and (2) global motion restoration that separates the

local motion from the global motion. In this method, each image is

considered as an ensemble of a number of blocks. Step (1) may employ

a full-searching block matching algorithm and results in a set of

motion vectors which capture the compound effect of zoom, pan, and

object movement. Step (2) groups the blocks which share the common

global motion parameters into one object. Three types of global

motion are considered. They are zoom, rotational pan and

translational pan. A single parameter for the entire image is

associated with each of the first two types of global motion and is

object independent. However, the motion vector introduced by pan

depends not only on the global pan vector but also depends on the

distance of object form camera. Therefore, in addition to the local

object movement, translational pan can be used for object

segmentation.

      The extraction of global motion parameters is accomplished via

least square estimation technique and singular value decomposition

(SVD). Once the global motion parameters are removed, images are

decomposed into objects of different movement. Thus, this method can

be used for moving object segmentation applications. Experimental

results show that this new technique is efficient in reducing the

entropy of the block motion vectors with zooming and panning

motion---up to an order of 30% to 40%. It can therefore be used to

increase the data compression efficiency.

 

[科資中心編號]RB8910-0383

[題    名]

影像壓縮、匹配、傳輸與加密(I)

Image Compression, Matching, Communication and Security (I)

[研 究 者 ]

鍾國亮

Chung, Kuo-Liang

[機構名稱  ]台灣科技大學資訊管理系(NTSTINM)

[經費來源  ]行政院國家科學委員會(NSC)

[補助金額  ]471000

[補助編號  ]NSC88-2213-E011-005

[研究開始日期]1998/08

[研究結束日期]1999/07

[語    文]中文

[建檔日期  ]2001/06

[出版情形  ]完成日期:民國88年12月

[原件提供  ]行政院國家科學委員會科學技術資料中心

NSC88-2213-E011-005

[科資分類號 ]IA0403; IA0300

[關 鍵 詞 ]

影像壓縮; 影像處理; 加密; 匹配; 傳輸

 

Image compression; Image processing; Encryption; Matching;

Transmission

 

[摘    要]

      影像處理這門領域一直不斷的發展著。人們對影像壓縮、匹配、影像傳輸和安全性等主題的關心程度有明顯的增加。這些主題也是目前主要的研究核心問題。

      計劃主持人同研究小組在過去幾年在這些方面累積不少經驗及成果。本研究小組提出此計劃並致力於其中之相關研究而且已獲得許多成果。目前有一項不同成果已被接受並將於近期刊登在國外著名期刊,以及有三項不同成果已投往國外著名期刊接受審查。本研究小組在今年度計畫中完成下列主要研究成果。(1)影像壓縮的成果[15]:先將霍夫曼樹轉換為單邊的霍夫曼樹,然後提出一個節省記憶體空間的有效資料結構表示法來表示此單邊的霍夫曼樹。此結果已被Theoretical Computer Science 接受並即將刊出。(2)影像匹配的成果 [16]:植基於均值金字塔的結構上,提出了一套快速的搜尋匹配演算法。此結果已投至IEEE Transactions on Image Processing接受審查。(3)影像在網路上的漸進式傳輸設計的成果[17]:在四元樹[5,6,11]壓縮技術的結構下,提出有效的資料分割演算法,及有效的資料儲存方式,以達到漸進傳輸的目的。此結果已在整理並準備投稿中。(4)影像在傳輸上的保密性研究的成果[18]:發展一套三階段式的影像加密技術,此技術兼具高壓縮率與高保密性的雙重效果。此結果已投至Journal of Visual Communication and Image Representation接受審查。

 

      The field of image processing is continually growing. It has

a significant increase in the level of interest in image

compression, matching, image communication, and security. These

four topics form the core of a major modernization. Our research

group has developed several efficient results and one of them is

accepted for publication on well-known international journal and

three are submitted to well-known international journal. In this

project, we do the research and have the results as shown below. (1)

A space-efficient huffman decoding algorithm and its parallelism is

obtained in the first part of our results and it was accepted for

publication on Theoretical Computer Science. (2) A faster search

algorithm for vector quantization based on mean pyramids structure

is obtained in the second part of our results and this part has

been submitted to IEEE Transactions on Image Processing for

possible publication. (3) Progressive image transmission based on

quadtree with shading approach is obtained in the third part of our

results. (4) Three-stage approach for encrypting images is obtained

in the forth part of our results and it has been submitted to

Journal of Visual Communication and Image Representation for

possible publication.

 

 

[科資中心編號]RB8806-0083

[題    名]

電子圖書館文獻與博物館藏品數位化技術與應用之研究---子計畫IV:電子圖書館

與博物館中影像處理(I)

Image Processing in Digital Library and Museum (I)

[研 究 者 ]

傅楸善

Fuh, Chiou-Shann

[機構名稱  ]台灣大學資訊工程系(NTUECSIE)

[經費來源  ]行政院國家科學委員會(NSC)

[補助金額  ]472000

[補助編號  ]NSC88-2213-E002-031

[研究開始日期]1998/08

[研究結束日期]1999/07

[頁 冊 數 ]4頁

[語    文]中文

[建檔日期  ]2000/04

[出版情形  ]完成日期:民國88年8月

[原件提供  ]行政院國家科學委員會科學技術資料中心

NSC88-2213-E002-031

[科資分類號 ]IA0403; IA0408

[關 鍵 詞 ]

影像處理; 數位圖書館; 數位博物館; 影像復原; 影像強化

 

Image processing; Digital library; Digital museum; Image

restoration; Image enhancement

 

[摘    要]

      本計畫為"電子圖書館文獻與博物館藏品數位化技術與應用之研究(II)"

之一分項計畫。旨在應用數位影像處理(Digital Image Processing)與電腦視

(Computer Vision)之各項方法,研發適合電子圖書館與博物館系統需求之影像提昇(Image Enhancement)與影像復原(Image Restoration)技術。電子圖書館與博物館系統為本研究群體計畫所選定之應用領域,故影像提昇與影像復原之相關技術必將扮演極重要的角色。本子計畫即針對上述兩項技術需求進行研發,不僅與其他子計畫有著密切的關係,更將是電子圖書館與博物館系統技術開發與實際應用成功的關鍵。

      影像提昇要處理影像使得它比原來影像更適合特殊應用,例如:當文件或

藏品由掃描器或數位相機輸入後,原件可能因為蛀蝕,水漬,剝落等原因已損壞,此時需要先將影像復原或提昇。其目的在於提昇影像品質,可理解性,可辨識性,視覺外觀等。影像提昇技術又可概分為空間範圍(Spatial Domain)與頻率範圍(Frequency Domain)方法。當使用者用電子圖書館系統在觀看資料或器物時,為了要看清楚某些特定的部分或強調某些特殊效果時,系統必須提供影像復原或提昇,以滿足使用者需求。

      影像復原要把退化或被雜訊敗壞的影像復原或重建,影像復原技術乃依

據退化模型(degradation model)應用反濾波方法(inverse filtering)來回復

原來影像。用數位相機輸入時,必然會有光學扭曲,尤其是大張地圖不能一次輸入時,需要接圖則一定要把光學扭曲修正回來,這是影像復原中很重要的工作。影像復原是數位影像處理中的重要技術,因為雜訊可能來自很多地方,例如:影像攝取時攝影機之CCD感應器雜訊,透鏡聚焦不良之模糊,傳送影像時之通訊雜訊,影像長期儲存之退化,物體因快速運動而模糊,傳送影像時編碼壓縮之品質損失...等。在這些狀況下都需要用到影像復原技術來復原或重建影像。

      本計畫將從事電子圖書館與博物館中影像提昇與影像復原技術之研究,

且推廣影像提昇與影像復原在電腦視覺,電腦自動檢測,高畫質電視,影像壓縮之應用。

 

      This proposal, "Image Processing in Digital Library and

Museum", is one of the subprojects of the "Research on Digitizing

Techniques and Applications of Digital Library Documents and Museum

Collections". The purpose is to apply various algorithms in digital

image processing and computer vision and develop image enhancement

and restoration techniques for digital library and museum. Image

enhancement and restoration and related processing play vital roles

in digital library and museum systems. This subproject is closely

related to other subprojects and is the key component for

successful digital library and museum research, techniques, and

applications.

      Image enhancement processes images to make it more suitable

than the original image for specific application. For example, when

documents or collections are input with scanner or digital camera,

image restoration or enhancement is needed due to rotting, water

damage, peeling, and so on. The purpose of image enhancement is to

improve image quality, intelligibility, visual appearances... etc.

Image enhancement techniques can be classified as spatial domain

and frequency domain methods. When a user uses digital library

system to browse documents or collections, the system must offer

image restoration or enhancement to satisfy user's requirement to

clarify a certain part or emphasize some special effect.

      Image restoration tries to restore or reconstruct degraded or

deteriorated images. Image restoration uses degradation model and

inverse filtering to restore images. Optical distortion always

exists when digital camera is used to capture documents or

collections. Optical distortion must be corrected especially when

large map cannot be input at a time and the parts must be mosaicked

back to the whole image. Image restoration is an important

technique in digital image processing. Noise can come from various

sources, e.g. CCD sensor noise during image acquisition, blurring

caused by improper focusing, noise in communication channel, image

degradation due to long-term storage, blurring caused by fast

object motion, quality loss incurred by image compression and

decompression...etc. Under all these situations, image restoration

is required to restore or reconstruct images.

      We will conduct research and development on various image

processing for digital library and museum and apply to computer

vision, automatic computer visual inspection, high-definition

television, and image compression.

 

[科資中心編號]RA77-2178

[題    名]

以多重決定做快速適應邊跡之尋找

The Multi-decision Concept for Adaptive Edge Detection.

[研 究 者 ]

林鎮藩;毛齊武

LAM,CHENG-FARN; MAO,CHI-WU

[機構名稱  ]成功大學電機工程系(NCKEELE)

[經費來源  ]行政院國家科學委員會(NSC)

[補助金額  ]0

[補助編號  ]NSC74-0611-E006-01

[研究開始日期]1985/01

[研究結束日期]1985/12

[出版日期  ]1986/03

[頁 冊 數 ]23

[語    文]中文

[建檔日期  ]1988/06

[科資分類號 ]IE;IE0510

[關 鍵 詞 ]

多重決定;邊跡;動態閥值;數位影像處理;可靠性

 

Multi-decision;Edge;Dynamic threshold;Digital image processing;Reliability

 

[摘    要]

   在本計畫中,我們發展出一種新的處理法則.根據該法則,我

們融會人類視覺對光強度適應性之影響參數,以及以多重決

定之方法做邊跡(edge)之探尋.由於邊跡是影像中的根本要素

之一,因此在數位影像處理中,如何自動尋找出所需之邊跡,就

成了一個主要的課題.對邊跡的探尋而言,在已有的許多方法

,其主要考慮的因素偏重於處理的速度和方法的可靠性.在

本計畫中,由於採用動態閥值(dynamic threshold)的新觀念,可自動

取得適宜的閥值,根據此一法則所得到的結果較一般常用的

算子更為接近人類主觀測試知(subjective test),使影像處理結果

之可靠性大為增加.

 

In this project, a new model of human visual system and an algorithm of mult

have been supplemented in the operator of edge detection. Since edges are cas

primitives of pictures, the detection of proper edges automatically has b the

main goals in digital image processing. So far as edge detections are cmany

edge recognizers have been formulated, most of which are concerned with of

processing and the reliability of the algorithm. In this research work,ept of

dynamic threshold has been adopted which is adaptive for the local arin a

proper threshold value automatically. The results thus obtained are com those

of Subjective Test [ 1] , and it is clearly shown that the edges obtaying this

model are more similar to subjective-test results than those produe other

popular operators.

 

科資中心編號: RA8410-2268 

題名: 智慧型微波透視子題三:微波透視識別之類神經系統建立

Intelligent Microwave Penetrative Vision: 3. The Design of Neural Network System for Microwave Penetrating Vision. 

研究者: 許超雲 ; 汪順祥

Hsu,Chau-Yun ; Wang,Shuenn-Chyang 

機構名稱: 大同工學院電機工程系(TTCTELE) 

經費來源: 行政院國家科學委員會(NSC) 

補助金額: 新台幣 1858 仟元 

補助編號: NSC81-0404-E036-08 

研究日期: 1991/08 - 1992/07 

出版日期: 1993/05 

頁冊數: 167頁 

科資分類號: IE0502;IA0403 

關鍵詞: 微波影像 ;Hopfield神經網路 ;倒傳遞模型 ;逆散射 ;影像重建

Microwave Image ;Hopfield Neural Network ;Back Propagation Model ;Inverse Scattering ;Image Reconstruction 

中文摘要: 在本子題中,我們利用倒遞模式和哈氏神經網路來解決埋藏物之微波影像重建。配合倒遞模式的應用,我們提出一非特定演算法來求取微波影像反散射問題之解。完全良導體之形狀可藉由所提神經計算模式之遞迴方法所獲得。在每一次遞迴的週期中,物體的形狀可經由能量函數的極小化來求得,此能量函數係計算真實散射場和利用遞迴式神經計算方法所求得之散射場之間的最小平方誤差。根據模擬結果我們可以發現我們所提出的神經計算模式能夠成功地解決微波影像之反散射問題。另外,配合哈氏神經網路的應用,微波影像重建的問題可以映射成一種最佳化問題;也就是說,我們必須定義一個適當的能量函數。根據此函數,我們建立一個具有零偏差輸入和自我相互連接加權的哈氏神經網路。由模擬的結果我們可以得到以下結論:(1)當埋藏物位於近場時,微波影像的效果較好。(2)對於好的重建效果,取樣距離是主要的關鍵,而非場平面中之取樣點。當場平面中之取樣距離大約為0.25至 0.50波長時,可得到很完美的微波影像。(3)若環境滿足(1)(2)時,可獲得完美的微波成像。 

英文摘要: In this subproject of Neural Group for this joint project in first year, we investigate with the problem that microwave image restoration of a buried object using a Back Propagation model (BP) and a discrete Hopfield Neural Network (HNN). To cooperate with BP, a nonalgorithm methods is presented for the solution of the inverse scattering problem of microwave image. The shape of the perfectly conducting object is obtained by iterative method with using the proposed neural computative scheme. During each iterative period, the shape solution of the object can be given by minimizing the energy function, least square error between the calculated and desired scattering field. Simulations show that the proposed neural computing method can solve the inverse scattering problem for microwave image successfully. To cooperate with HNN, the problem of microwave image restoration is mapped into an optimization problem; i.e., it is needed to define a suitable energy function. Based on this function, we construct a HNN which possesses zero bias input and self- interconnection weight. The simulation results conclude (1) the restoration of microwave image perform well when the buried object are in near field. (2) The sample points in field plane is not a key for good restoration, but the sample distance is. When the sample distance in field plane is about 0.25 to 0.5 wavelength, the performance is near perfect. (3) If a environment satisfy (1) and (2) a perfect microwave image could be obtained. 

 

 

[科資中心編號]RA8501-2053

[題    名]

應用卡門濾波器的影像編碼

Image Coding by Kalman Filtering.

[研 究 者 ]

周本生

CHOW,BEN-SHUNG

[機構名稱  ]中山大學電機工程系(NSYEELE)

[經費來源  ]行政院國家科學委員會(NSC)

[補助金額  ]309000

[補助編號  ]NSC80-0404-E110-07

[研究開始日期]1990/08

[研究結束日期]1992/03

[出版日期  ]1993/10

[頁 冊 數 ]56頁

[語    文]英文

[建檔日期  ]1996/10

[出版情形  ]研究報告,56頁,民國82年10月

Report, 56p., Oct. 1993

[原件提供  ]行政院國家科學委員會科學技術資料中心

NSC80-0404-E110-007

[科資分類號 ]IA0403;IE0508

[關 鍵 詞 ]

影像壓縮;卡門濾波器;影像結構;影像重建;壓

縮比

 

Image compression;Kalman filter;Image texture;Image reconstruction;

Compression ratio

 

[摘    要]

      卡門濾波器已被廣泛應用於影像還原領

,本計畫是將卡門濾波器之應用推展於影像

壓縮,影像壓縮的解碼部分在觀念上與影像還

原相似,但編碼部分卻與影像還原的對應部分

---影像生成模式大為不同,這也說明了卡門濾

波器在影像壓縮的發展受到阻礙的原因。

      在影像還原領域內,影像生成模式是建立

於一大批同性質影像的統計知識上,並且資訊

壓縮的目的並未受到考慮,在本計畫影像編碼

的壓縮考量下,我們介紹一種影像結構的觀念,

它是影像鄰方居點在亮度值上彼此的一種線性

關係。影像亮度的變化可能很大,但影像結構

即彼此的關係卻變化和緩而適於壓縮。

      卡門濾波器的估測是建立在其信號生成模

式與資料觀察模式上,所以此二種模式對於我

們的壓縮編碼法之建立極為重要,我們是以影

像結構來決定信號生成模式,而以部分取樣為

資料觀察模式,因此部分取樣的多寡是我們壓

縮率的瓶頸,對於某些應用如與現行電視相容

的高畫質電視之壓縮,部分取樣的資料是原本

有在不必考量在壓縮範圍內的,本計畫的方法

便很適合推展應用,最重要的是本計畫提出一

種新的編碼觀念其應用更有待發展。

 

      Kalman filter has been well applied to the field of image

restoration. In this project, we extend its application to image

coding. The process of image decoding is similar in concept to image

restoration. However, the concept of image modeling, a prerequisite

procedure for image restoration, is very much different from the

concept of image encoding. This may be the reason for the inhibition

of Kalman filtering's application to image coding.

      In image restoration, the image is modeled on the statistical

knowledge of an ensemble of images. Moreover, the compression

requirement is not considered. For the purpose of image encoding, we

introduce the concept of image texture to model the image. Image is

defined to be the linear relation between the brightness of

neighboring pixels. Brightness function usually varies very much but

the texture function can be smooth enough for compression.

      The signal model and observation models are essential to

Kalman filtering and thus our coding method as well. Image texture

governs the system parameters of our signal model. The observation

model is set up by sub-sampled image, which is thus the bottleneck

of our compression ratio. However, for some applications such as

high definition television the sub-sample image is available and not

considered for compression if compatibility to present TV system is

required. Most importantly, the novelty of our method of image

coding still opens for new application.

 

 

[科資中心編號]RA81-2773

[題    名]

運用二階KALMAN濾波器作影像修復

Image Restoration Using 2-Order Kalman Filter.

[研 究 者 ]

張克章

CHANG,KER-CHANG

[機構名稱  ]國防管理學院資源管理研究所(NDMGRSMG)

[經費來源  ]行政院國家科學委員會(NSC)

[補助金額  ]415000

[補助編號  ]NSC80-0404-E123-01

[研究開始日期]1991/02

[研究結束日期]1992/01

[出版日期  ]1992/02

[頁 冊 數 ]57

[語    文]中文

[建檔日期  ]1994/05

[出版情形  ]中華民國八十年電腦視覺、圖學

暨影像處理研討會論文集,pp.59-65, 1991

[科資分類號 ]IE;IE0304

[關 鍵 詞 ]

二階卡門濾波器;影像還原;添加性雜訊; 灰暗度

 

Second-order Kalman filter;Image restoration;Additive noise;Gray value

 

[摘    要]

      本研究最主要的目的,為針對遭受添加性 雜訊干擾之影

像提出一套新的修復技術,稱之 為一維二階影像修復模組,本

模組乃根據傳統 一階卡門濾波器模組擴增而成。

      此技術係假設影像上每一列,其各光點之 灰暗度皆和其

前二光點有密切的關聯性,進而 利用二階卡門濾波器求出每

一光點灰暗度之最 小均方差估計值。本計畫認為如果有更

多有用 的資訊從鄰近光點獲得,其修復品質應可較一 階卡門

濾波器為佳。

      實驗結果顯示,本研究方法所得之修復影 像中,其雜訊去

除效果比較一階濾波器模組有 明顯的改進,達到影像修復品

質提昇之目的。

 

      The purpose of the research suggests a new restoration technique for

images with additive noise.

      The proposed technique called one-dimensional second-order image

restoration model, is based on the second-order Kalman filter system which is

an extension from the conventional first-order Kalman model. The proposed

technique treats the gray value of a pixel on each line of the image in a way

that it has a significant correlation with two previously processed pixels. A

minimum mean squared error estimation of each pixel is derived using the

second-order Kalman filter. The authors believe that while the first-order

Kalman filter is feasible to the image restoration problem, an improved

quality of restoration should be found if more available information extracted

from the neighbored pixels can be applied. Experimental results of several

testings verify our idea that the proposed technique dose provide better

restoration quality than that can be done by the first-order model.

 

[科資中心編號]RA81-2918

[題    名]

視覺模式併合之影像還原系統設計

Design of Visual Model Combined Image Restoration System.

[研 究 者 ]

溫瓖岸

WEN,KUEI-ANN

[機構名稱  ]交通大學電子研究所(NCTUELNG)

[經費來源  ]行政院國家科學委員會(NSC)

[補助金額  ]399000

[補助編號  ]NSC81-0404-E009-132

[研究開始日期]1991/08

[研究結束日期]1992/07

[出版日期  ]1992/08

[頁 冊 數 ]72

[語    文]英文

[建檔日期  ]1994/05

[科資分類號 ]IE;IE0303

[關 鍵 詞 ]

影像還原;即時處理;VLSI設計;視覺模式

 

Image restoration;Real-time processing;VLSI design;Visual model

 

[摘    要]

      本計畫完成一系列嶄新的影像處理系統體 系,它是以人

類的視覺模式為基礎,並利用模糊 集合的觀念做為視訊的評

估。考慮在人類視覺 系統中對於背景亮度和空間頻率的敏

感性,併 合濾波和遮罩處理,建立一個非線性視力臨界 空間

模式,經由此模式失真量測所產生訓練數 據的反應,藉以估計

定性的診斷策略,並利用模 糊一般性規則與關聯性推論方法,

模擬高階視 訊中央處理系統。

      此一視覺模式成功的應用在影像品質評斷 、影像壓縮

和影像強化處理,經由視覺模式的 加入,其處理效能較一般影

像處理系統,在主觀 的視覺品質上有著顯著的提昇。

 

      In this project a series of novel processing schemes for gray level

image processing systems based on the visual model and using the concept of

fuzzy sets is proposed. The spatial model of the threshold vision that

incorporates the filtering and masking processes takes account of two major

sensitivities of the human visual system, namely background illumination

levels and spatial frequency sensitivities. The distortion measures use common

sense fuzzy rules for image quality prediction and the diagnosing strategy of

each phase of the response is obtained by estimating qualitative but not

quantitative fuzzy rules from the training data generated by the visual model.

This fuzzy associated reasoning tries to imitate high level visual central

processing.

      Three areas in which human visual models have been successfully applied

are image quality assessment, image compression, and image enhancement. By the addition of the visual model, performance of these systems have a visible

improvement of subject quality depending on visual perception.

 

[科資中心編號]RA8404-2180

[題    名]

高畫質訊號處理研究(III)---子計畫(五)影像復 原技術於HDTV取

樣率轉換之應用

Application of Image Restoration Techniques in HDTV Sampling Rate Conversion.

[研 究 者 ]

曹建和

TSAO,JEN-HO

[機構名稱  ]台灣大學電機工程系(NTUEELE)

[經費來源  ]行政院國家科學委員會(NSC)

[補助金額  ]212100

[補助編號  ]NSC83-0404-E002-010

[研究開始日期]1993/08

[研究結束日期]1994/07

[出版日期  ]1994/09

[頁 冊 數 ]53

[語    文]中文

[建檔日期  ]1996/01

[科資分類號 ]IE;IE0508;IA0403

[關 鍵 詞 ]

影像重建;高清晰度電視;取樣率轉換

 

Image reconstruction;High definition television;Sampling rate conversion

 

[摘    要]

      應用數值方法分析流經VLCC(Very large cruel- oil carrier)的外

流場是本文研究的目的。分析的 重點主要在於探討艉跡流

場的速度變化與沿船 身表面的壓力分布。以期能進一步掌

握艉部較 佳的線型及對行進中的船隻所產生的影響。

      業已於第一年內建立了一套能分析三維、 時間平均、

不可壓縮的Navier-Stokes方程組的程 式,並驗證了此一程式的正

確性。此一方法是 採用有限容積方法,其中待解的速度因變

數為 一經過非正交網格轉換後的逆變速度分量。非 線性的

流動項是以QUICK的方法處理。經離散後 的代數式將以SIMPLEC

的模式迭代地求得收斂解 。在本年度中的主要工作是將自

由液面的邊界 條件加入上述的三維程式,並以Wigley-hull船型

為 測試例子。

 

      Application of a finite volume method to analyze the external flow field

over a VLCC is the main purpose of the present study. In the first year, we

have established a computer code (formulation) to analyze a set of

three-dimensional, time-averaged, incompressible Navier-Stokes equations.

      The main work accomplished in the first year of contract period is

summarized in this paragraph. The basic equations represented by the

contravariant velocities were discretized by a finite volume method where the

nonlinear convection terms were approximated by a QUICK scheme. The resulting algebraic equations were solved in a decoupled manner by the SIMPLEC iterative solution algorithm. In the present (second) contract year, we furthermore incorporate the free surface boundary condition into the verified 3D

Navier-Stokes code. We closed Wigley-hull form as the working target because

its simplicity to avoid any consequence due to the complexity of the VLCC ship

form.

 

[科資中心編號]RA8504-2085

[題    名]

同時的影像認定與回復

Simultaneous Image Identification and Restoration.

[研 究 者 ]

賴坤財

LAY,KUEN-TSAIR

[機構名稱  ]台灣工業技術學院電子工程技術系

(NTITENT)

[經費來源  ]行政院國家科學委員會(NSC)

[補助金額  ]341000

[補助編號  ]NSC83-0404-E011-015

[研究開始日期]1993/08

[研究結束日期]1994/07

[出版日期  ]1994/07

[頁 冊 數 ]80頁

[語    文]中文

[建檔日期  ]1997/01

[出版情形  ]研究報告,80頁,民國83年7月

Report, 80p., Jul. 1994

[原件提供  ]行政院國家科學委員會科學技術資料中心

NSC83-0404-E011-015

[科資分類號 ]IA0403

[關 鍵 詞 ]

多通道影像;影像回復;模糊認定;交互通道相關

 

Multichannel image;Image restoration;Blur identification;Cross-

channel correlation

 

[摘    要]

      使用頻率領域的多通道濾波器回復受損的

彩色影像時,可以利用交互通道的相關性和交

互通道模糊的資訊,使得回復的影像較以單通

道影像回復法回復的影像為佳,這是多通道影

像回復法的最大優點。而在計算量上,因以多

通道影像回復法回復彩色影像,只需處理3*3的

矩陣,相對於單通道的影像回復法,並沒有增加

太多的計算量。在本計畫中,我們以虛擬反向

濾波器及溫納濾波器作影像回復,因為溫納濾

波器需要使用影像頻譜,我們討論數種擴充為

多通道二維頻譜估計的頻譜估計法。

      與影像回復密切相關的問題為模糊之認定

。到目前為止,文獻上探討模糊認定的方法,大

部分只考慮單通道的情形。單通道影像處理的

方法並沒有辦法對交互通道模糊作認定。我們

使用期望值最大化(Expectation-maximiaztion,簡稱EM)

演算法代替最大相似性(Maximum likelihood,簡稱ML)

參數估計法,解決非線性的最大化問題。在頻

率領域,多通道的方法同樣只需要處理3*3的矩

,便能對受損的彩色影像同時作認定和回復

 

      In this research, multi-channel (MC) image identification and

restoration are addressed. In MC image restoration, cross-channel

correlations and cross-channel blurs are exploited to achieve a

better restored image than the result from single-channel (SC)

restoration filters. Computationwise, MC restoration increases the

complexity only slightly because 3*3 matrices are the major entities

encounted in the computation. In this research, pseudo-inverse and

Wiener filters are tested. Various MC spectral estimation techniques

are used to obtain the image spectrum for the use in the Wiener

filter.

      In image identification, The blur and noise statistics that

caused the degradation of an image is to be estimated. So far the

research on image identification found in the literature has been

exclusively on single-channel cases. In this research, we try to do

MC image identification, where cross-channel blurs are part of the

parameters to be estimated. We apply the EM (expectation-

maximization) algorithm to compute the ML (maximum-likelihood)

estimates of the relevant parameters. In the frequency domain, the

MC identification only needs to process 3*3 matrices, which is

simple in computation.

 

[科資中心編號]RB8701-0313

[題    名]

類神經網路與模糊理論應用於水下聲學影像識別之研究

Underwater Imagery Pattern Recognition Using Neural Nets and Fuzzy

Theory.

[研 究 者 ]

王榮華

Wang, Jung-Hua

[機構名稱  ]台灣海洋大學電機工程系(NTOSELE)

[經費來源  ]行政院國家科學委員會(NSC)

[補助金額  ]273000

[補助編號  ]NSC85-2611-E019-032

[研究開始日期]1995/08

[研究結束日期]1996/07

[頁 冊 數 ]120頁

[語    文]中文

[建檔日期  ]1998/10

[出版情形  ]完成日期:民國85年8月

[原件提供  ]行政院國家科學委員會科學技術資料中心

NSC85-2611-E019-032

[科資分類號 ]IG0104

[關 鍵 詞 ]

可適性模糊中值濾波器;影像復原;影像識別;類神經網路;智慧型機械手臂

 

Adaptive fuzzy median filter;Image restoration;Imagery recognition;

Neural network;Intelligent manipulator

 

[摘    要]

      本計畫研究將類神經網與模糊理論結合,發展出最佳化適應性模糊濾波器(Adaptive Fuzzy Optimal Filter:AFOF)及適應性模糊倒傳遞混成網路(Adaptive Fuzzy Backpropagation Hybrid Network:AFBHN),以應用於雜訊影像

的概化還原上。

      其中,AFOF的Optimal表示網路除了取最佳的像素值(from

defuzification)當輸出外,也表示網路經由學習後,得到一組"optimal"

membership functions以還原雜訊影像。AFOF彌補了傳統Median Filter(MF)在還原高impulsive noise影像(>30%)時,效果不彰的缺點,且其在處理低雜訊比(<20%)影像上,也獲致比其他濾波器更好的效果。在濾除Gaussian white noise方面,AFOF對不同variance下的高斯雜訊影像作還原,其效能也優於WienerFilter。另外,本研究所發展出來的AFBHN,其系統中的倒傳遞網路,不但對影像提供了誤差修正的功能,更提昇AFOF在高雜訊(.gtoreq.30%)時的還原能力;而對於Gaussian white noise,還原效果更勝於AFOF及Wiener Filter。

      在概化(Generalization)還原應用上, AFOF及AFBHN只要學習同性質環境下的一張影像,便可建構好此同類影像的還原網路。日後,對於同性質環境下產生的雜訊影像,都能做有效率的復原;特別當雜訊增大時,效果愈優於傳統方法(MF及WF)。此外AFBHN在高斯雜訊方面,其概化還原的能力,不僅增進了AFOF在此方面的表現,更突顯了此混成網路在濾除此種雜訊時的優點。在水下聲學影像部分,限於水下設備的不足及人力配合等問題(本計畫原先提出時為一整合計畫之子計畫),且因濾波還原為影像辨識成敗之關鍵,因此本計畫之研究今年先集中研究陸上影像之濾波還原相關技術。最後,誠摯感謝國科會之支持使本計畫得以順利完成。

 

      In this project, we combine neural networks and fuzzy theory to

propose two nonlinear filters for image restoration with learning

generalization capability, namely adaptive fuzzy optimal filter (AFOF)

and adaptive fuzzy backpropagation hybrid network (AFBHN).

      The AFOF not only overcomes the weakness of conventional median

filter (MF) in the case of large impulsive noise rates, but also

improves the performance of other filters in the case that the

impulsive noise rate is below 20%. The AFBHN restores noisy images

even when the impulsive noise rate is over 30%. Under Gaussian white

noise testing, the AFBHN has shown better results than the Wiener

filter (WF) and AFOF.

      As to generalization capability, the AFOF and AFBHN can be

trained by one image randomly chosen from a set of similar images.

After training, they can efficiently restore the rest similar images

degraded by either impulsive noise (specially in the high impulsive

noise rate) or Gaussian white noise of up to variance=0.2. Under

Gaussian white noise, the AFBHN improves the ability of AFOF and

provides better filtering performance than the wiener filters.

      We sincerely thank NSC for their generous support on this

project. Finally, we regret that due to lacking underwater equipment

in our lab, we were forced to conduct ordinary image restoration

techniques that are essential to the future success of underwater

research.

 

[科資中心編號]RB8702-0289

[題    名]

交談式智慧型系統於遙控載具的應用---即時的影像和數據通訊信

Intelligence Interactive Neural Fuzzy System for Remote Vehicle

Applications-Real Time Image and Data Communication.

[研 究 者 ]

陳嘉琳

CHEN, CHIA-LIN

[機構名稱  ]中華工學院資訊工程系(CHPICOS)

[經費來源  ]行政院國家科學委員會(NSC)

[補助金額  ]288700

[補助編號  ]NSC85-2612-E216-001

[研究開始日期]1995/08

[研究結束日期]1996/07

[頁 冊 數 ]45頁

[語    文]中文

[建檔日期  ]1998/12

[出版情形  ]完成日期:民國85年12月

[原件提供  ]行政院國家科學委員會科學技術資料中心

NSC85-2612-E216-001

[科資分類號 ]IE0507;IF1306;IA0403

[關 鍵 詞 ]

數據通信;影像壓縮;無線通訊;即時;影像復原;立體影像處理;區域網路;資料加

 

Data communication;Image compression;Wireless communication;Real time;

Image restoration;Stereo image processing;Local area network;Data encryption

 

[摘    要]

      本計畫是一實做性質的計畫,所要執行的是:遙控載具與監視站主機之間之通信管道雛型的建立。此管道雛型的建立有助於即時的影像和數據通信,以做為交談式智慧型系統中,人機界面訊息傳遞的通信管道。本計畫結合了影像的壓縮、計算機區域網路通信等技術,是一個具整合性質的研究題材。在此計畫中,我們將探討現有之影像處理、影像壓縮及網路通信方法,並分析探討這些技術在此計畫上的適用性。我們在IBM PC上進行「點對點之有線通信網路及界面的建立」,使用Winsock套件來撰寫TCP/IP protocol以控制網路卡,傳輸資料,以模擬載具與監視站之間的通信,在研究成果上,在Window視窗下所發展的影像通信環境雛型,已建立完成,而且成效上大致不錯,將來可繼續在功能上擴展。此外,我們使用「失真的」影像壓縮方法MPEG,來處理需要即時呈像的視訊影像;對用於爾後影像分析、不需即時處理的影像,則研究「不失真的」影像壓縮方法(LJPG),以保有原來影像的所有訊息,從實驗中得知,LJPG不失真的影像壓縮法的壓縮率相當不錯。將來的工作是延伸此計畫的研究成果並整合各項的功能。

 

      The goal is project is to establish a real-time communication

channel for image and data transmission, which can be further

utilized as a channel for information exchange for the ultimate human-

machine interface for the interactive intelligent system. This

project integrates many techniques such as image and video

compression, computer networking and so on. In this project, we study

on the existing techniques of lossless image compression, video

compression and TCP/IP networking, and investigate the applicability

of these techniques to the goal of this project. We implement "point-

to-point communication and interface" based on the Winsock toolkit

for TCP/IP protocol so as to control the networking cards and

transmit data. As shown in the experiments in this report, we have

built prototype of this communication channel. The lossless

compression method used is call LJPG which also shows good

compression ratio. The video compression is entirely based on MPEG

standard. The future work will focus on integrates the techniques

described above in one system.

 

[科資中心編號]RB8704-0008

[題    名]

類神經網路與模糊理論應用於影像識別之研究

A Study of Intelligent Underwater Manipulator System---Pattern

Recognition Based on Neural NETS and Fuzzy Theory

[研 究 者 ]

王榮華;詹寶珠

Wang, Jung-Hua;Chung, Pau-Choo

[機構名稱  ]台灣海洋大學電機工程系(NTOSELE)

[經費來源  ]行政院國家科學委員會(NSC)

[補助金額  ]180000

[補助編號  ]NSC86-2611-E019-024

[研究開始日期]1996/08

[研究結束日期]1997/07

[頁 冊 數 ]60頁

[語    文]英文

[出版情形  ]完成日期:民國86年8月

[原件提供  ]行政院國家科學委員會科學技術資料中心

NSC86-2611-E019-024

[科資分類號 ]IA0407;IE0601;IA0406

[關 鍵 詞 ]

圖形辨識;類神經網路;模糊理論;水下機器臂;中值濾波器;適應性濾波器;影像還

;歸屬函數

 

Pattern recognition;Neural network;Fuzzy theory;Underwater

manipulator;Median filter;Adaptive filter;Image restoration;

Membership function

 

[摘    要]

      本子計畫係研究應用神經網路及模糊理論於影像處識別,並發展其模擬及應用軟體,以期應用於整合型計畫---智慧型水下機械臂系統。

      一年來本子計畫已成功發展一套可應用於還原脈衝雜訊影像之適應性神經模糊濾波器(Histogram-based adaptive neuro-fuzzy filter,HANF)。利用輸入影像之像素分布統計,我們提出可以估測出一組最佳(Well-conditioning)初始隸屬函數(HIMF)之演算法。若與傳統隨機選定初始隸屬函數法比較,經吾人以實驗證明利用HIMF訓練HANF,可以獲致更快之收斂速度(因訓練時僅須微調)及較佳之影像還原效果(因HIMF在誤差曲面上較易避開區域性最小值)。其實,若估測之像素分布統計與原始像素分布誤差不大,則單以HIMF隸屬函數即足以獲得近似最佳還原效果。以PSNR (Peak signal-to-noise ratio)作評比指標,HANF在脈衝雜訊率>20%情況下表現均比傳統中值濾波器好。本文亦針對像素分布統計與最佳初始隸屬函數之關連性作理論探討及實驗,以瞭解HANF的概汎能力及適應性質。吾人認為由於HANF具良好的概汎能力以及適應性,足以減緩甚或免除"再訓練"之困擾,在未來實際應用(如:多媒體影音傳輸)上相當具有可行性。

      另外,本子計畫亦研發一改良式中值濾波器。此一新的中值濾波器可以毋需訓練,徹底免除"再訓練"之困擾之。實驗結果證實改良式中值濾波器無論在高SNR或低SNR環境下,均比傳統中值濾波器或HANF有較佳之影像還原效果。

 

      This paper develops a novel adaptive filter, namely, Histogram-

based Adaptive Neuro-fuzzy Filter (HANF) powerful for restoring

images corrupted by impulsive noise. Effective training HANF is

achieved by using an algorithm based on histogram statistics to

estimate a set of well-conditioning initial membership functions

(IMF). HANF is capable of fast learning because only fine-tuning is

needed to complete the training process. Contrast to the inefficient

random strategy, the histogram-inferred IMF significantly reduces the

likelihood of converging to an undesirable local minimum in the error

surface. In cases where training is prohibited, the histogram-

inferred IMF alone can achieve near-optimal restoration. We show that

HANF is superior over traditional median filter (MF) in PSNR (peak

signal-to-noise ratio) performance, especially when impulsive noise

rate >20%.

      Furthermore, correlation between histogram statistics and HIMF

is explored to study adaptive property and generalization capability

of HANF. With the histogram-inferred IMF, HANF essentially is free

from the retraining problem. Comparisons of various experimental

results are given to show superiority of HANF in restoration

performance and in practical implementation feasibility.

      A new algorithm incorporated with standard median filter is

proposed to effectively remove impulsive noise in image processing.

This computationally efficient approach first classifies input pixels

and then perform median filtering process. Simulation results show

that the new scheme, regardless of high SNR or low SNR, displays

superior mean square error (MSE) over plain MF and HANF.

 

[科資中心編號]RB8805-0125

[題    名]

基於類神經網路及模糊理論之智慧型濾波系統之架構設計與其在影像復原上之應用

On the Design of Neuro-Fuzzy Intelligent Filtering System with the

Application on Image Restoration

[研 究 者 ]

游寶達

Yu, Pao-Ta

[機構名稱  ]中正大學資訊工程系(CHCUCSIE)

[經費來源  ]行政院國家科學委員會(NSC)

[補助金額  ]362500

[補助編號  ]NSC87-2213-E194-021

[研究開始日期]1997/08

[研究結束日期]1998/07

[頁 冊 數 ]4頁

[語    文]中文

[建檔日期  ]1999/11

[出版情形  ]完成日期:民國87年11月

[原件提供  ]行政院國家科學委員會科學技術資料中心

NSC87-2213-E194-021

[科資分類號 ]IA0403; IA0407

[關 鍵 詞 ]

濾波系統; 影像復原; 模糊理論; 類神經網路; 類神經學習法則; 模糊迴歸分

; 適應性; 強勢性

 

Filtering system; Image restoration; Fuzzy theory; Neural network;

Neural learning algorithm; Fuzzy regression analysis; Adaptation;

Robustness

 

[摘    要]

      在本計畫中,共分四大研究論題。在第一個論題中,我們依據著名的

Sugeno's fuzzy controller提出一個基於模糊知識庫的濾波系統,本系統利用

三條模糊法則及三種傳統濾波器,並輔以類神經網路學習能力使其記住多種高低雜訊的知識,有效地還原多種高低汙染的影像。在第二個論題中,我們提出一個具學習能力的前置濾波器簡稱RONN濾波器,其最大的特色是利用一個前置器稱impulse noise detector的輔助來篩選training signals,此功能大大的提高RONN濾波器學習能力(即記住高低雜訊的知識)並藉此增加RONN濾波器之適性與韌性的能力。在第三個論題中,我們提出一個植基於適性模糊線性回歸的錯誤剪除中值濾波器。此濾波器是最先利用模糊迴歸分析的理論所設計出。其利用一個adaptive fuzzy regression neural selector的輔助來提升RCRS濾波器的濾波效能。從大量的實驗中可明確地觀察出本計畫提出三個濾波器之濾波效能超越其他主流的濾波器。

      在第四部分中,我們將本計畫所設計出來的系統,局部的功能已經以JAVA建構於本實驗室的網站,其網址為http://neuralpc5.cs.ccu.edu.tw/lab315/paper.html。

 

      In this report, four main parts of research results are

presented. In Part I, on the basis of Sugeno's fuzzy controller, we

propose a new class of fuzzy filters, which uses three fuzzy rules

and three kinds of conventional filters for improving the

performance of conventional filters. Moreover, the neural-network

techniques are employed to train the fuzzy rules for memorizing the

noise knowledge such that our fuzzy filters are effectively able to

restore the corrupted images with different probabilities of noises.

In Part II, we propose a new class of nonlinear filters called rank

ordered neural network (RONN) filters based on a

detection-estimation strategy and neural network technique. RONN

filters can select the training signals to increase their learning

ability. This can further improve the adaptive and robust filtering

capabilities of RONN filters. In Part III, we propose a novel class

of nonlinear filters called the error-trimmed median (ETM) filters.

This is a first paper to propose the filtering technique based on

fuzzy regression theory. ETM filters exploit the adaptive fuzzy

regression neural selector and incorporate with RCRS filters to

achieve noise cancellation effectively.

      In Part IV, some of functions of the RONN and ETM filters are

implemented by JAVA. The URL of web site of our laboratory http:

//neuralpc5.cs.ccu.edu.tw/lab315/paper.html.

 

[科資中心編號]RB8805-0152

[題    名]

馬可夫隨機場在影像處理上最佳化問題之研究

The Study of the Optimization on Image Restoration Based on Markov

Random Field

[研 究 者 ]

張國華

Change, Kuo-Hwa

[機構名稱  ]中原大學工業工程系(CYUEINE)

[經費來源  ]行政院國家科學委員會(NSC)

[補助金額  ]379400

[補助編號  ]NSC87-2213-E033-008

[研究開始日期]1997/08

[研究結束日期]1998/07

[頁 冊 數 ]7頁

[語    文]中文

[建檔日期  ]1999/11

[出版情形  ]完成日期:民國87年10月

[原件提供  ]行政院國家科學委員會科學技術資料中心

NSC87-2213-E033-008

[科資分類號 ]IB0206

[關 鍵 詞 ]

馬可夫隨機場; 影像處理; 最佳化; 馬可夫決策過程

 

Markov random field (MRF); Image processing; Optimization; Markov

decision process

 

[摘    要]

      在一張大小為N*N具有256灰階之數位影像上,我們可將看成一個N*N個代表灰階值之隨機變數,我們稱為隨機場。數位影像在傳輸過程中,會受到雜訊之干擾而失真。我們將收到已受干擾之影像當作已知,配合隨機場之機率模式,以貝氏估計量(Bayesian estimator)估計真實(無雜訊)之影像。亦即,我們以在給定干擾影像之下,使條件機率最大之最佳解,為估計之真實影像。我們以作業研究之方法去探討其最佳化問題並利用模擬退火法及吉伯斯抽樣(Gibbssampler)求得還原動態影像。另外配合時間序列分析,分析影像中之數據資料,而求出符合之適當之參數。

 

      We restore a sequence of noised dynamic image. We use the

noised image as a given condition, and by coupling with the Markov

random field (MRF) as a prior, we obtain the estimated real image

by maximizing the posterior. In our study, we use MRF to model the

image, use Gibbs sampler to updating the gray level, and use

simulated annealing to maximize the posterior. We also use time

series to estimate the proper parameters during our maximization.

 

[科資中心編號]RB8902-0042

[題    名]

線性移動模糊影像之參數估測演算法之設計

Design of Estimating Motion Parameters for a Linear Motion Blurred

Image

[研 究 者 ]

陳永盛

Chen, Yung-Sheng

[機構名稱  ]元智大學電機工程系(YZUELE)

[經費來源  ]行政院國家科學委員會(NSC)

[補助金額  ]403000

[補助編號  ]NSC88-2213-E155-023

[研究開始日期]1998/08

[研究結束日期]1999/07

[頁 冊 數 ]5頁

[語    文]中文

[建檔日期  ]2000/10

[出版情形  ]完成日期:民國88年12月

[原件提供  ]行政院國家科學委員會科學技術資料中心

NSC88-2213-E155-023

[科資分類號 ]IA0403

[關 鍵 詞 ]

影像處理; 線性移動模糊影像; 影像復原

 

Image processing; Linear motion blurred image; Image restoration

 

[摘    要]

      根據人類視覺系統對移動影像擁有之基本知覺特性,本計畫提出的重點

在於建立一可行之線性移動估測器,其可以估測移動方向與移動量。我們提出之作法概分為兩個部分。第一部分包括:轉換模糊影像至角樣型之二值影像、將影像分割微小區域、並在每一小區域中使用Hough detection的方法把角度資訊抽取出來、最後產生整個影像之角度資訊分布曲線,便可將移動方向估測出來。第二部分包括:使用影像強化技術將模糊影像之移動量資訊強化出來、使用四分樹(分割-合併)法將影像之活動區域抽取出來(此活動區域為含有明顯移動量資訊之區域)、建立一計算公式其可以計算影像整體移動量的分數、建立一映成函數(根據移動視知覺特性)使得所計算之分數可以對應出影像移動量。本計畫以商用影像處理軟體對一些常用之影像產生不同之線性移動模糊影像,以進行實驗分析及討論。

 

      In this project, based on the human visual perception, we

design a feasible approach to simplify the problem of estimating

the motion parameters (motion direction and motion extension) for a

linear motion blurred image. Two main parts of the proposed

approach are as follows. The first part includes: a scheme

developed to transform the given blurred image into an

angle-pattern binary image; a Hough transform adopted to extract

the angle information in each subregion; and an angle-information

distribution constructed to detect the motion direction. The second

part includes: a scheme developed for the enhancement of the given

linear motion blurred image; a quadtree method adopted to

split-and-merge the activity regions containing the motion

extension information; a formula constructed to compute the total

score for the image; and a logarithmic regression mapping function

constructed to estimate the motion extension. The various linear

blurred cases of different often-used images were produced by an

image processing tool (e.g., PHOTO-SHOP software) and used in our

experiments.

 

科資中心編號: RB9010-0146 

題名: 智慧型通訊影像處理系統之設計、分析與應用

On the Design and Analysis of Intelligent Image Processing and Its Applications under the Communication Environment 

研究者: 游寶達

Yu, Pao-Ta 

機構名稱: 中正大學資訊工程研究所 

經費來源: 行政院國家科學委員會 

補助金額: 新台幣 757 仟元 

補助編號: NSC89-2213-E194-019 

研究日期: 1999/08 - 2000/07 

出版日期: 2000/08 

頁冊數: 4頁 

原件提供: 行政院國家科學委員會科學技術資料中心, NSC89-2213-E194-019 

科資分類號: IA0403 

關鍵詞: 類神經網路 ;浮水印 ;資料隱藏 ;模糊系統 ;遺傳演算法 ;影像復原 ;彩色影像

Neural Network ;Watermark ;Data Hiding ;Fuzzy System ;Genetic Algorithm ;Image Restoration ;Color Image 

中文摘要: 在本計畫中,我們提出一種新的建構在類神經學習網路的數位浮水印技術,來對彩色圖片做所有權辨認。我們的方法是將一張暗藏的浮水印藏入我們所要保護的彩色圖片之中,並且利用類神經網路來有效的學習我們所藏入的浮水印和加入浮水印之後的圖形之間特有的關係。如此之技術有助於通訊影像的私密性及安全性。由於類神經網路學習的程序和適應性的能力,所以訓練過後的類神經網路幾乎可以確實的從加入浮水印的彩色圖形中取出所嵌入的浮水印,並且能夠對抗一般影像處理的攻擊。從許多的實驗結果中,我們可以看出我們的方法對於一些影像處理和攻擊是非常強韌的。 

英文摘要: In this report, we propose a novel digital watermarking technique based on neural networks for color images. Our technique hides an invisible watermark into a color image, and then effectively employ neural networks to learn the characteristics of the embedded watermark related to the watermarked image. Due to neural networks possessing the learning and adaptive capabilities, the trained neural networks almost exactly recover the watermark from the watermarked image against image processing attacks. Extensive experimental results illustrate that our technique significantly possesses robustness to be immune against the attacks. 

 

科資中心編號: RB9101-0518 

題名: 數位影像還原技術在刑事鑑識工作上之研究與應用

Digital Image Restoration Technology and Its Application to Forensic Science 

研究者: 溫哲彥

Wen, Che-Yen 

機構名稱: 中央警察大學鑑識科學系 

經費來源: 行政院國家科學委員會 

補助金額: 新台幣 359 仟元 

補助編號: NSC89-2218-E015-002 

研究日期: 2000/08 - 2001/07 

出版日期: 2001/08 

頁冊數: 4頁 

原件提供: 行政院國家科學委員會科學技術資料中心, NSC89-2218-E015-002 

科資分類號: IA0403;SE0813 

關鍵詞: 影像還原 ;三維轉換 ;刑事鑑識 ;影像資料 ;指紋 ;犯罪問題 ;刑事科學

Image Restoration ;Three Dimensional Transformation ;Forensic Identification ;Image Data ;Fingerprint ;Crime Problem ;Forensic Science 

中文摘要: 在刑事鑑識的工作中,隨著影像科技的進步,由犯罪現場所獲得的影像資料(例如指紋、現場照片與監視錄影帶等)已有愈來愈多的趨勢,然而這些資料往往由於犯罪現場或多或少被破壞、取樣儀器的限制或人為操作等等的因素,常常並不完整或不清楚(或遭現場背景雜訊干擾),因而造成了明明有記錄,但是卻變成了無效證據,甚為可惜。1950 年代,美國與前蘇聯由於太空計畫中的需要而開始了影像還原技術的研究,隨後這項技術亦被應用於醫學影像與電影娛樂界等其他用途。近年來,在刑事鑑識界亦開始嘗試應用這項技術來還原模糊的監視錄影帶、車牌照片等常見的影像鑑識問題。然而,由於影像還原技術在應用時,與所要面對的問題或對象有密切的關係,而且目前現有的還原方法均有其適用範圍與限制,加上許多刑事鑑識工作上常遇到的案件均具有結合數種雜訊與PointSpread Function(PSF)因素所造成的影響,如果只單純使用目前現有影像還原技術,其效果並不如預期。本計畫提出應用3D 轉換的方式將歪斜車牌還原並正規化,以利車牌之辨識。 

英文摘要: In the forensic science field, there are many criminal data including fingerprints, scene photos and surveillance camera videotapes. The evidence may become the key point to solve crime problems. However, the evidence may be contaminated by the noise from crime scenes, or distorted by improper collected. In this case, the evidence will become void and we may lose the chance to arrest the criminals. The field of image restoration began primarily with the efforts of scientists involved in the space programs of both the United States and the former Soviet Union in the 1950s and early 1960s. The image restoration technology was also extended to other fields, such as the medical imaging and movie media. Recently, some researchers tried to apply this technology to analyze data from the crime scenes. However, because current image restoration technologies are case-dependent, we cannot obtain efficient results when applying them to forensic cases directly. In this project, a novel three-dimensional transformation method for vehicle license plate number normalization is proposed. This method provides an efficient solution to normalize skew distortion vehicle license plate images. The Hough transform is used to estimate the license plate position and the rotation angle needed to normalize a vehicle license plate image. After the three-dimensional transformation and normalization, the vehicle license plate numbers are recognized easily. 

 

科資中心編號: RB9111-0070 

題名: 動態影像序列的解析和恢復

Frame to Frame Integration and Restoration for Dynamic Imagery 

研究者: 潘敏政

Pan, Min-Cheng 

機構名稱: 東南技術學院電子工程系 

經費來源: 行政院國家科學委員會 

補助金額: 新台幣 311 仟元 

補助編號: NSC90-2213-E236-004 

研究日期: 2001/08 - 2002/07 

出版日期: 2002/09 

頁冊數: 6頁 

原件提供: 行政院國家科學委員會科學技術資料中心, NSC90-2213-E236-004 

科資分類號: IA0403 

關鍵詞: 動態影像序列 ;影像整合 ;影像還原 ;空間解析度 ;框架

Dynamic Image Sequence ;Image Integration ;Image Restoration ;Spatial Resolution ;Frame 

中文摘要: 本計畫,針對所獲得的影像序列畫質並不是很理想甚或模糊不清,進行動態影像序列的解析和恢復技術。首先,將一般最常用的類比訊號(訊號記錄於磁帶上,如VHS系統)轉換為數位影像訊號;然後,將此動態數位影像序列進行解析,可獲得無雜訊的背景影像資訊,藉此可將移動物予以抽離或重置於此較清晰的背景影像;然後對有興趣的部分進行影像恢復。第一點,應用超解析技術以恢復影像的空間頻率的高頻部分;第二點,發展新技術以增加影像的空間解析度以能達到進一步辨識的可能性。 

英文摘要: This research will study the techniques of frame to frame integration and restoration to process a series of images acquired from an analog imaging system. To improve qualities of these frames, the following steps are adopted: (1) frame to frame integration for dynamic imagery: first, analog image signals recorded on the tape of a VHS system are to be transformed into digital ones; then, multi- frame integration could be performed to obtain a noiseless background image which moving objects can be segmented from or replaced on; (2) frame to frame restoration: if originally blurred by defocus or motion blur, such processed images from previous steps will be restored using super-resolution algorithms to improve spatial frequencies of these image frames; moreover, this research will develop a novel technique to improve spatial resolution of image of interest for image recognition. 

 

科資中心編號: RN9102-0148 

題名: 移動模糊影像的恢復和超解析技術之研究

Study of Restoration and Super-Resolution Techniques for Motion Blurred Images 

研究者: 潘敏政 

機構名稱: 東南技術學院電子工程系 

經費來源: 行政院國家科學委員會 

補助編號: NSC89-2218-E236-001 

研究日期: 2000/08 - 2001/07 

出版日期: 2001/10 

頁冊數: 10頁  

原件提供: 行政院國家科學委員會科學技術資料中心, NSC89-2218-E236-001 

科資分類號: IA0403 

關鍵詞: 影像處理 ;影像復原 ;影像解析度

Image Processing ;Image Restoration ;Image Resolution 

中文摘要: 本計畫將研究影像恢復和超解析三方面課題:第一、組裝一維線性運動模擬器,第二、建立各類型的運動模型,包含位移不變性的線擴展函數﹝如等、加、減速運動﹞與位移相依性的線擴展函數﹝如焦距擴展或收縮運動﹞,第三、應用影像恢復演算法或超解析技術恢復以上所提的運動模糊影像。 

英文摘要: In this project, there are three aspects to be studied for restoration and super-resolution: (1) the set-up of the 1-D liner motion simulator; (2) various prototypes of motion including constant velocity, acceleration, deceleration which result in space-invariant line spread function (SILSF) as well as the focus of expansion or contraction which result in space-variant LSF (SVLSF); and (3) the restoration of motion blurred images taking from the real scenes to validate the proposed methods. 

 

科資中心編號: RB9101-0557 

題名: 適用於衛星影像壓縮之高效率影像切割技術之研究

An Efficient Image Segmentation Technique for Satellite Image Compression 

研究者: 張麗娜

Chang, Le-Na 

機構名稱: 台灣海洋大學商船系 

經費來源: 行政院國家科學委員會 

補助金額: 新台幣 312 仟元 

補助編號: NSC89-2611-E019-063 

研究日期: 2000/08 - 2001/07 

出版日期: 2001/07 

頁冊數: 4頁 

原件提供: 行政院國家科學委員會科學技術資料中心, NSC89-2611-E019-063 

科資分類號: IG0001 

關鍵詞: 影像壓縮 ;衛星影像 ;影像切割 ;電腦模擬

Image Compression ;Satellite Image ;Image Segmentation ;Computer Simulation 

中文摘要: 本研究提出一適用於衛星影像壓縮之高效率影像切割技術。此技術充分利用影像在頻帶間及空間資料之關連性。首先,依據衛星影像之資料自相關矩陣所對應之主要特徵向量定義一影像區塊結合之判斷準則,依此準則將具有相同的區域特性之影像切割為適當的特徵區間。接著,對各區間影像進行對應之KL轉換,得一頻帶間已去相關性之特徵影像。最後,再以JPEG進行空間資料壓縮。電腦模擬驗證此壓縮技術具有極佳之壓縮效能。 

英文摘要: In the study, we present an effective segmentation technique for multispectral image compression. This technique fully exploits the spectral and spatial correlation in the data. The original image is first divided into some proper eigen-regions according to the local terrain characteristics of image. Then, each region image is transformed by the corresponding KL transformation function and results in an eigen-region image for further compression. Simulation tests performed on Landsat TM images have demonstrated that the proposed compression scheme is suitable for multispectral image. 

 

 

科資中心編號: RB9101-0013 

題名: 總計畫

MPEG-4 Multimedia Technology 

研究者: 杭學鳴 ; 王聖智 ; 張文鐘 ; 林大衛 ; 蔣迪豪

Hang, Hsueh-Ming ; Wang, Sheng-Jyh ; Chang, Wen-Thong ; Lin, David Wdarwei ; Chiang, Ti-Hao 

機構名稱: 交通大學電子工程系 

經費來源: 行政院國家科學委員會 

補助金額: 新台幣 1668 仟元 

補助編號: NSC89-2213-E009-241 

研究日期: 2000/08 - 2001/07 

出版日期: 2001/09 

頁冊數: 5頁 

原件提供: 行政院國家科學委員會科學技術資料中心, NSC89-2213-E009-241 

科資分類號: IA0403 

關鍵詞: 多媒體通訊 ;視訊編碼 ;影像切割 ;多媒體資料庫

Multimedia Communication ;Video Coding ;Image Segmentation ;Multimedia Database ;Mpeg-4 ;Mpeg-7 

中文摘要: 過去十五年間,多媒體通訊的研究以驚人的速度成長。本計畫的目標在於研究並實現國際標準MPEG-4中幾項重要的技術。MPEG-4標準頗為龐大複雜,本整合型計畫將就其中之視訊處理、視訊資料編碼與傳輸以及系統等三大部分予以研究,共分五項子計畫:(1).MPEG-4相關之彩色影像切割技術研究;(2).利用適應性維納濾波器於低位元率視訊訊號之量化效應的消除;(3).比例式視訊編碼技術及視訊通訊終端機技術之研究;(4).MPEG-4與MPEG-7系統之研究;(5).多媒體架構與數位視訊浮水印在網際網路之應用。 

英文摘要: Researches in the area of multimedia communications have been advancing at an amazing speed in the past 15 years. As a result, the standardization of H.261/H.263, JPEG and MPEG series have produced a number of very popular commercial products in the market. On the other hand, the standard activities stimulate and push the advances of researches in this area. The goal of this project is to study and/or implement several important technologies in the MPEG4 standards. The MPEG-4 standard is rather huge and complicated. This project focuses on three major parts of an MPEG-4 system: (1).video data processing,(2).video data coding/transmission and (3).system issues. We will also extend our researches beyond MPEG-4. Particularly, the MPEG-7 and MPEG-21 systems will also be investigated. This project consists of 5 sub-projects: (1).Study of MPEG4-related image segmentation techniques; (2).Quantization effect reduction with adaptive Wiener filter in low bit rate video; (3).Research in scalable video coding techniques and visual communication terminal technologies; (4).A study on MPEG-4 and MPEG-7 systems; (5).Multimedia framework and digital video water-marking for the Internet applications. 

 

 

科資中心編號: RB9101-0011 

題名: 子計畫II:MPEG-4 相關之彩色影像切割技術研究

Study of MPEG4-related Image Segmentation Techniques 

研究者: 王聖智

Wang, Sheng-Jyh 

機構名稱: 交通大學電子工程系 

經費來源: 行政院國家科學委員會 

補助金額: 新台幣 580 仟元 

補助編號: NSC89-2213-E009-231 

研究日期: 2000/08 - 2001/07 

出版日期: 2001/08 

頁冊數: 4頁 

原件提供: 行政院國家科學委員會科學技術資料中心, NSC89-2213-E009-231 

科資分類號: IA0403 

關鍵詞: 影像切割 ;色彩空間

Image Segmentation ;Mpeg-4 ;Color Space 

中文摘要: 在本年度的計畫中,我們討論了現有影像切割演算法的關鍵問題,並利用「資料趨勢」發展了一套新的彩色影像切割演算法。整個研究計畫內容包含四個部分:色彩空間的選擇、現有影像切割方法的問題討論、影像資料趨勢的擷取及影像切割演算法的設計。在色彩空間的選擇上,我們採用一個類似人類視覺系統中互補色系統的色彩空間;在現有影像切割方法的問題討論上,我們對於影像切割提出了有異於傳統的判斷方式;在影像資料趨勢的擷取上,我們以類似DOG運算子的方式設計了一個可以擷取出影像亮度趨勢與色彩趨勢的運算子來;在影像切割方法的設計上,我們發展了一個新的彩色影像切割演算法,此演算法在準確度與穩定性上都較現有演算法為佳。 

英文摘要: In this project, we discuss the key problems in current image segmentation algorithms and utilize "data trend" instead of "data variation" to develop a new algorithm for color image segmentation. This project includes four major parts: selection of color space, discussion of current image segmentation algorithms, detection of data trend, and the proposed algorithm. In the selection of color space, a color space obeying the opponent-colors principle is adopted. In the discussion of current image segmentation algorithms, we propose a new criterion for segmentation judgment. In the detection of data trend, an operator is developed to extract image data trend. In the development of segmentation algorithm, a new color segmentation algorithm is proposed to provide better accuracy and stability. 

 

 

[科資中心編號]RB8910-0328

[題    名]

以物件為基礎之視訊壓縮演算法之研究

The Research of Coding Algorithms for Object-Based Video Compression

[研 究 者 ]

陳美娟; 鄭博允

Chen, Mei-Juan; Cheng, Po-Yuen

[機構名稱  ]東華大學電機工程研究所(NDHUELEG)

[經費來源  ]行政院國家科學委員會(NSC)

[補助金額  ]420000

[補助編號  ]NSC88-2218-E259-002

[研究開始日期]1998/08

[研究結束日期]1999/07

[語    文]中文

[建檔日期  ]2001/06

[出版情形  ]完成日期:民國88年12月

[原件提供  ]行政院國家科學委員會科學技術資料中心

NSC88-2218-E259-002

[科資分類號 ]IA0403; IA0101

[關 鍵 詞 ]

視訊壓縮; 視訊編碼; 動態影像專家群; 物件導向; 影像分割; 形狀編碼; 移

動估測; 多媒體

 

Video compression; Video coding; Moving picture experts group

(MPEG); Object orientation; Image segmentation; Shape coding;

Motion estimation; Multimedia

 

[摘    要]

      數位視訊產品及服務目前正以驚人的速度在民生電子、電腦以及通訊工業上積極發展中。在影像視訊儲存與傳輸上,工業界訂定的標準如早期的JPEG,JBIG,MPEG-1,H.261到MPEG-2,H.263等,開啟了多媒體與數位電視的時代。      MPEG-4是ISO組織建立的一個有效率的編碼標準,除了融合了MPEG-1 和MPEG-2 影像編碼功能以外,並首先加入"以內容為基礎"的視訊編碼方式,其採用以物件為基礎的壓縮方式,影像畫面被分割成一些任意形狀的物件平面,並針對物件進行形狀編碼、移動估計與紋理編碼。形狀編碼的目的是為了得到較好的影像品質與增加編碼效益,任意形狀物件的編碼成敗將影響影像重建品質與編碼的位元率。本研究根據物件形狀,利用極座標領域提出減少形狀點數的技術,並於傳統形狀編碼方式做進一步的改良,達到在無失真或低位元率編碼上更佳的壓縮效率與影像品質。

 

      MPEG-4 standard will allow the transmission of the

arbitrarily shaped video object. It allows the encoding of video

objects using motion, texture and shape information. In this

project, a new vertex-based shape coding in polar coordinates is

proposed. To optimize the coding efficiency of object boundary, a

new vertex selection scheme and the area distortion criterion are

presented. The proposed vertex selection scheme can reduce the

shape points of the boundary for lossless coding. For lossy coding,

the proposed area distortion criterion provides good

rate-distortion trading-off.

 

[科資中心編號]RB8910-0262

[題    名]

物件導向視訊媒體壓縮技術之研發

Object-Oriented Video Data Compression Techniques

[研 究 者 ]

吳家麟

Wu, Ja-Ling

[機構名稱  ]台灣大學資訊工程系(NTUECSIE)

[經費來源  ]行政院國家科學委員會(NSC)

[補助金額  ]545000

[補助編號  ]NSC88-2213-E002-051

[研究開始日期]1998/08

[研究結束日期]1999/07

[語    文]中文

[建檔日期  ]2001/06

[出版情形  ]完成日期:民國88年12月

[原件提供  ]行政院國家科學委員會科學技術資料中心

NSC88-2213-E002-051

[科資分類號 ]IA0403

[關 鍵 詞 ]

視訊壓縮; 物件導向; 動態影像專家群; 影像分割; 互動性

 

Video compression; Object orientation; Moving picture experts group

(MPEG); Image segmentation; Interactivity

 

[摘    要]

      隨著使用者對'互動性'要求的提高,新一代的MPEG壓縮標準:MPEG-4對壓縮基元的定義已由畫面(Frame)轉換成物件(Object)。本計畫即針對此一趨勢,研討以物件為基礎之視訊壓縮技術。影像分割技術,物件特徵壓縮技術及物件紋理壓縮技術將是我們探討的主要課題。此外,為使參與計畫者對MPEG-4標準有更深一層的體認,我們以Java實作出一MPEG-4解碼器,其中"高互動性"是此解碼器之特點。

 

      Along with the high demand of "interactivity" between users

and media, MPEG-4 (the new member of MPEG family) extends to

delivery of more complex, structured experiences combining the

immediacy of audio/video with downloaded A/V object structures that

a user might change with interaction. Based on this observation,

this project focused on the development of "object-oriented" data

compression techniques for video, which includes: Image

segmentation, object feature compression, and object texture

compression as its major components. For recognizing the in-depth

of the MPEG-4 standard, a highly interactive MPEG-4 decoder has

also been constructed in this project.

 

[科資中心編號]RB8901-0022

[題    名]

適用於MPEG4視訊資料之影像切割技術

Image Segmentation Techniques for MPEG4 Video

[研 究 者 ]

洪一平

Hung, Yi-Ping

[機構名稱  ]中央研究院資訊科學研究所(ASII)

[經費來源  ]行政院國家科學委員會(NSC)

[補助金額  ]404000

[補助編號  ]NSC88-2213-E001-018

[研究開始日期]1998/08

[研究結束日期]1999/07

[頁 冊 數 ]6頁

[語    文]中文

[建檔日期  ]2000/06

[出版情形  ]完成日期:民國88年7月

[原件提供  ]行政院國家科學委員會科學技術資料中心

NSC88-2213-E001-018

[科資分類號 ]IA0403

[關 鍵 詞 ]

影像切割; 視訊壓縮; 動態影像專家群; 動態影像分析

 

Image segmentation; Video compression; Moving picture expert group

(MPEG); Dynamic image analysis

 

[摘    要]

      為使視訊系統具有物件導向的處理能力,首先必須解決視訊物體(Video

object)切割的問題,也就是如何從序列影像中分辨出不同的視訊物體。在這個計畫中,我們提出一個名為River的新技術用於切割視訊物體。這個技術可以分為兩個階段:分析階段以及切割階段。在分析階段,River先將一段截取的視訊資料轉換成許多的flow-threads,然後將每個flow-thread分別計算出一組特徵,稱為ft-features。所謂的flow-thread是經由追蹤序列影像中像素的運動,而產生一連串相對應像素的集合。經由flow-thread中像素間的運動,便可以計算出多種適合視訊物體切割的統計資料做為ft-features。在接下來的切割階段,利用ft-features和傳統的Pattern Classification技術,可以將flow-threads區分出不同的視訊物體。和其他已知的分割技術比較,River的優勢在於不是採用初始分割,再加上物體追蹤的方式,卻同時地考慮所有序列影像中的運動。River的另一個優點,它可以讓使用者很容易地經由多媒體系統,以互動的方式來觀看及修飾切割的結果。為了使River切割出的輪廓更加精準,我們採用Snake技術做進一步的輪廓修正。

 

      In order to let a video processing system have the

object-oriented functionality, it is important to first solve the

problem of video object (VO) segmentation, or more precisely, the

problem of extracting VOs from an image sequence. In this project,

we propose a novel technique, called river, for VO segmentation.

The river technique can be divided into two phases: the analysis

phase and the segmentation phase. In the analysis phase, the river

technique first converts a finite duration of image sequence into a

dense field of flow-threads and then extracts a set of features

from each flow-thread. Here, a flow-thread is an ordered list of

pixels obtained by tracing the motion vectors along the image

sequence. Given a flow-thread, some flow-thread features (or

ft-features) can be extracted for the purpose of VO segmentation.

Next, in the segmentation phase, the flow-threads are grouped into

several VOs by applying conventional pattern classification

techniques---based on the ft-features. Compared to the existing VO

segmentation methods, the river technique has the advantage that it

does not decompose the VO segmentation problem into an initial

segmentation problem (segmenting a single image frame) followed by

a temporal tracking problem. Instead, motion information contained

in the whole image sequence is taken into consideration

simultaneously. Another advantage of the river technique is that

the ft-features can be easily perceived and utilized by the user of

an interactive multimedia system. In order to obtain more accurate

contours of the video objects, we use the snake technique to refine

the contours.

 

[科資中心編號]RB8805-0131

[題    名]

模組式模糊類神經網路於影像分割之應用

A Modular Fuzzy Neural Network for Image Segmentation

[研 究 者 ]

詹寶珠

Chung, Pau-Choo

[機構名稱  ]成功大學電機工程系(NCKEELE)

[經費來源  ]行政院國家科學委員會(NSC)

[補助金額  ]424000

[補助編號  ]NSC87-2213-E006-023

[研究開始日期]1997/08

[研究結束日期]1998/07

[頁 冊 數 ]30頁

[語    文]英文

[建檔日期  ]1999/11

[出版情形  ]完成日期:民國88年4月

[原件提供  ]行政院國家科學委員會科學技術資料中心

NSC87-2213-E006-023

[科資分類號 ]IA0403; IA0407

[關 鍵 詞 ]

影像分割; 電腦斷層; 類神經網路; 模糊理論; 前後文資訊; 模糊規則; 腹部

解剖

 

Image segmentation; Computed tomography (CT); Neural network; Fuzzy

theory; Contextual information; Fuzzy rule; Abdominal anatomy

 

      Recognizing abdominal organs is one of essential steps in

visualizing organ structure, for providing assistant in teaching,

clinic training, and diagnosis. This paper describes a framework

for automatic abdominal organ recognition from a series of CT image

slices, designed based on shape analysis, image contextual

constraint, and between-slice relationship. Two processing phases,

feature extraction (object segmentation) and recognition, are

included in this framework. In the phase of object segmentation, a

multimodule contextual neural network is applied to segment each

image slice into disconnected regions. For each region, its shape

features, including relative location, relative distance, tissue

intensities, area sizes, compactness, and elongatedness are

calculated, along with its spatial relationships with respect to

spine. Then, according to the knowledge of anatomy, these features

are constructed to form fuzzy rules used for organ recognition. In

the recognition phase, the obtained features and the overlapping

information between adjacent slices are used for identifying each

organ. In this phase, a contour modification mechanism is also

embedded to adjust the abnormal extrusions caused by the partial

volume effect, and at the same time, smooth the organ's contour.

This proposed method has been tested on many clinic patient cases.

Results indicate that this method can successfully recognize

abdominal organs, not being affected by partial volume effects.

 

[科資中心編號]RB8805-0114

[題    名]

最佳化影像分割使用位元率及曲解度定理

The Optimal Quadtree Segmentation Using Rate-Distortion Theory

[研 究 者 ]

張隆紋

Chang, Long-Wen

[機構名稱  ]清華大學資訊科學系(NTHUCOS)

[經費來源  ]行政院國家科學委員會(NSC)

[補助金額  ]491800

[補助編號  ]NSC87-2213-E007-006

[研究開始日期]1997/08

[研究結束日期]1998/07

[頁 冊 數 ]4頁

[語    文]中文

[建檔日期  ]1999/11

[出版情形  ]完成日期:民國87年11月

[原件提供  ]行政院國家科學委員會科學技術資料中心

NSC87-2213-E007-006

[科資分類號 ]IA0403; IA0401

[關 鍵 詞 ]

影像分割; 最佳位元分配法; 失真; 灰階

 

Image segmentation; Optimal bit allocation; Distortion; Gray level

 

[摘    要]

      本研究計畫植基於位元率-曲解度定理(Rate-Distortion),提出一最佳

化之四元樹影像分割技術來壓縮灰階靜態影像。在於有限的位元數內,求得一最佳化四元樹影像分割方式,使得重建後的影像之平方誤差最接近原始影像。此類限制之最佳化問題可藉由拉格朗日乘子法(Lagrange Multiplier)來結合位元率及曲解度而變成相等的無限制的最佳化問題。我們可以藉由快速的多邊形包搜尋法(convex hull bisection algorithm)來求得最佳的拉格朗日乘子,並進而求得最佳化的四元樹影像分割。本研究計畫所提出的最佳化影像分割法可配合向量量化(Vector Quantization),離散餘弦轉換(Discrete Cosine Transform),方塊截斷編碼(Block Truncation Coding)。亦可應用於凌波轉換後之係數分割,預期本研究計畫所提出之分割方式應用非常廣泛且能達到非常高的效能,在理論及應用上均甚為重要。

 

      In this proposal, we propose an optimal quadtree segmentation

based on rate-distortion theory to compress still images, such that

the distortion of the reconstructed image is minimal under the

given bit budget constrain. That is, the reconstructed image has

the highest PSNR and the number of bits don't exceed the bit budget.

The above constraint optimal problem can be converted to an easy

equal unconstrained problem by merging the bitrate and distortion

by a Lagrange multiplier. The technique of our proposed quadtree

segmentation can be applied in many block-based compression schemes,

such as variable blocksize vector quantization (VBVQ), variable

blocksize DCT (VBDCT) or variable blocksize block truncation (VBTC).

It can also code the coefficients of the wavelet transform. The

optimal quadtree segmentation can achieve higher performance than

those use the fixed blocksize coding schemes.

 

 

[科資中心編號]RB8707-0356

[題    名]

強調空間連接性的影像分割

Image Segmentation with Emphasis on Spatial Adjacency

[研 究 者 ]

周本生

Chow, Ben-Shung

[機構名稱  ]中山大學電機工程系(NCSEELE)

[經費來源  ]行政院國家科學委員會(NSC)

[補助金額  ]187000

[補助編號  ]NSC86-2213-E110-030

[研究開始日期]1996/08

[研究結束日期]1997/07

[頁 冊 數 ]40頁

[語    文]中文

[建檔日期  ]1999/10

[出版情形  ]完成日期:民國86年8月

[原件提供  ]行政院國家科學委員會科學技術資料中心

NSC86-2213-E110-030

[科資分類號 ]IA0403

[關 鍵 詞 ]

影像分割;影像結構;區域擴張;相連接區域標記法

 

Image segmentation;Image structure;Region growing;Connected component

labeling

 

[摘    要]

      本計畫提出一個建立在影像結構的性質函數,並結合相連物件標記法之新影像分割方法。基於影像結構概念,定義了新的性質函數---像點間線性關係方程組,其包含性質相近和空間鄰近的資訊,滿足影像分割的兩大要求。

      影像分割是計算機視覺應用中的一項重要技術。其實現上有兩大難題:第一、不易選擇具有均勻特性的性質,做為分割區域之判斷依據。第二、分割出形狀的區域,其內相鄰像點的連接問題。詳細的說,區域內的像點有兩大考慮要素一、性質相近;二、空間鄰近。性質相近,是區域內應具有均勻的特性,而區域和區域之間,則須有明顯分別;空間鄰近,是影像分割必備的要素。只有相鄰近的像點或區域,才會考慮是否連接或合併為同一區域。這兩大要素是影像分割的重要觀念,也是我們發展方法所依循的原則。

      影像結構是定義於像點和左、上鄰居點亮度函數的線性關係。也就是用像點間亮度函數的線性關係方程式,可以來決定其影像結構的特性。一般影像分割都以純量做為性質,我們以線性方程組為性質函數,隱含抽象的概念,無法直接度量。但卻可藉由量測區域和像點合併的誤差,來判斷影像結構的一致性。

      相連接區域標記法原發展於黑白影像,藉由我們提出的性質函數,而擴展至灰階影像的應用上。基於處理的效率考量,我們引入Run-length標記法,作為演算法的基本架構。在計算性質函數的誤差和標記出分割區域,都是以上而下、左而右的掃描順序。總而言之,由於遞迴式的計算架構和掃描方式,將使我們發展的影像分割演算法,非常有效率。並且新性質函數的提出,成功地掌握空間連接和性質相近的特性,而能達到分割物件的好效果。

 

      A new image segmentation algorithm based on the textural

information of image pixels is presented. Image segmentation is a

fundamental technique for the application of computer vision. There

are two difficulties for the technique of image segmentation. First,

the choice of the characteristic with which the regions of an image

segmentation are homogeneous enough for the decision. Second, the

connection problem for pixels within the segmented components in an

arbitrary shape. In this thesis, we propose a new property vector

related to the concept of image texture and employ the technique of

connected components labeling to solve the problems.

      The image structure are defined to be the linear relationship

between pixels with their upper and left neighbors. Image can be very

inhomogeneous. However, the image structure may be uniform enough in

some image components. By this image texture, individual image

constraint equations can be set up for those triple pixel. These

equations are the property vectors chosen by us for segmentation

decision.

      Connected components labeling method was originally developed

for the binary images. By the introduction of our property vector,

the labeling method are extended into the gray image segmentation.

For computing efficiency, the Ronse and Devijver's run-length version

of labeling method is modified by us for our application. Our methods

of computing segmentation error for property vectors and labeling

segmentation components are both based upon a top-down and left-right

scanning order. As a summary, our segmentation computing are very

efficient due to the recursive computation structure and the scanning

method.

 

[科資中心編號]RB8705-0141

[題    名]

ISODATA法進行MR腦部影像之組織分類及精確功能影像定位研究

MR Brain Tissue Segmentation and Accurate Registration of Functional

Image Using ISODATA

[研 究 者 ]

林康平

Lin, Kang-Ping

[機構名稱  ]中原大學電機工程系(CYUEELE)

[經費來源  ]行政院國家科學委員會(NSC)

[補助金額  ]325000

[補助編號  ]NSC86-2213-E033-035

[研究開始日期]1996/08

[研究結束日期]1997/07

[頁 冊 數 ]41頁

[語    文]中文

[出版情形  ]完成日期:民國86年8月

[原件提供  ]行政院國家科學委員會科學技術資料中心

NSC86-2213-E033-035

[科資分類號 ]IA0403;IA0406;IA0407

[關 鍵 詞 ]

影像分割;模糊聚類法;模糊C均數聚類法;K均數聚類法

 

Image segmentation;Fuzzy clustering method;Fuzzy C-means clustering

method;K-means clustering method

 

[摘    要]

      本研究提出以ISODATA聚類法,並與K-Means(KM)、Fuzzy C-Means(FCM)及馬可夫隨機場(MARKOV RANDOM FIELD,MRF)等四種影像聚類方法,分別針對含有高雜訊之數位化醫學影像進行分割處理實驗與應用。

      在影像分割模擬實驗,ISODATA等四種聚類方法,在分割含有50%高雜訊的訊號,聚類點數誤差最佳可達3.23%。

      在影像分割處理應用上,以ISODATA聚類法對PET醫學影像進行分割處理,再利用其所得之分割結果,實際應用於自動量測hypoxia fraction參數,及MR影像與PET影像資料整合上。這些應用都獲得良好的結果,驗證所提四種方法對影像之分割均具可行性。

 

      In this thesis, some experiments and applications for

processing high noise medical images using segmentation methods are

presented. For two-dimensional image segmentation, four clustering

methods that are K-Means (KM), Fuzzy C-Means (FCM), ISODATA and

Markov random field (MRF) are exploited to compare their results and

performances.

      In simulated image, noise levels from 10% to 50% were added.

Among the four proposed methods, they provided the best image

segmentation result which gave 3.23% MSE in segmenting the image with

50% noise level.

      In applications, the same procedure was applied to measure

automatically the contours of MR and PET images. The ISODATA image

segmentation technique was also applied to PET image to automatically

measure the hypoxia fraction of nasopharyngeal carcinoma (NPC)

patient. The result of segmented PET image can be integrated with its

MR image for the visualization.

 

[科資中心編號]RB8705-0119

[題    名]

神經網路動態立體視覺系統之研究

An Active Stereo Vision System Based on Neural Approach

[研 究 者 ]

王榮華;詹寶珠

Wang, Jung-Hua;Chung, Pao-Choo

[機構名稱  ]台灣海洋大學電機工程系(NTOSELE)

[經費來源  ]行政院國家科學委員會(NSC)

[補助金額  ]315000

[補助編號  ]NSC86-2213-E019-006

[研究開始日期]1996/08

[研究結束日期]1997/07

[頁 冊 數 ]67頁

[語    文]英文

[出版情形  ]完成日期:民國86年8月

[原件提供  ]行政院國家科學委員會科學技術資料中心

NSC86-2213-E019-006

[科資分類號 ]IA0403

[關 鍵 詞 ]

立體視覺;類神經網路;深度圖;影像分割

 

Stereo vision;Neural network;Range map;Image segmentation

 

[摘    要]

      本計畫研究建立一神經網路動態立體視覺系統(Active stereo vision

system)。此系統除了能主動的控制攝影機之角度,使得欲觀察之區域能出現於兩攝影機之影像上,並使用不同類型神經網路來處理影像分割、完成Stereomatching的相關運算步驟,最後建立此系統可見範圍的精確深度圖。

      首先,為改善影像分割之效果,我們發展完成一個神經網路的自我產生競爭學習演算法,稱做週期活動力守恆(PVC)。我們賦予每個神經元一個活動力指標,以估測其在競爭學習中贏和輸的機率。此活動力指標充分能反映輸入資料之統計特性,使得網路能週期性的增加或移除神經元。在任何時刻,網路均維持活動力總和為1,稱之為活動力守恆。這個守恆原則最大優點是幫助我們推導出重要學習參數的設定。增加或移除神經元會影響輸入向量空間之分割,也改變了網路的熵值。由於PVC嘗試最大化其熵值,我們稱PVC是一個近最佳解的向量量化器。PVC的自我產生以及可調學習率之機制使得PVC能兼顧網路的學習能力與穩定性,這也讓PVC能在批次或是線上操作都能有很好的表現。在影像編碼的實驗中,我們比較了PVC和三種有名演算法---Kohonen self-organizing feature map(SOFM)〔2〕,FSCL〔7〕和SCONN2〔6〕的效能;模擬結果顯示出PVC比其他方法之優越處。      我們進一步將PVC運用在一個以神經網路為基礎的立體視覺系統中。藉由Sobel之運算得到立體影像每一點之特徵,如灰階值,變化量和方向性。再將這些特徵給PVC進行叢集影像;相同的特徵也用來訓練倒傳遞網路成為一個匹配器。訓練後的網路匹配器可用來產生一個初始的深度圖。之後,再使用一個以影像叢集為基礎的對應點演算法,以得到一個更精確的深度圖。其中,我們使用了一些限制來縮小搜尋空間並決定對應點,如Epipolar line,Ordering,Geometric和Local-

support等。最後的實驗結果也顯示出倒傳遞網路匹配器和對應點演算法的成效

 

      A self-creating neural network effective in learning vector

quantization, called PVC (Periodical Vitality Conservation) is

developed. Each neuron in PVC is characterized by a measure of

vitality. Conservation is achieved by bounding the summed vitality of

all neurons at constant 1, as vitality values of all neurons are

updated after each input presentation. We show that PVC effectively

fulfills the conscience principle and achieves biologically plausible

self-development capability. In addition, conservation in vitality

facilitates systematic derivations of learning parameters, including

the adaptive learning rate control useful in accelerating the

convergence as well as in improving node-utilization. The performance

of PVC is compared with three famous algorithms---Kohonen's Self-

Organizing Feature Map (SOFM) [ 1 ] , FSCL [ 2 ] , SCONN2 [ 3 ] ..

      The PVC is further used in a stereo vision system based on

neural networks [ 4 ] . Sobel operators are used to extract features

of intensity, variation, and orientation from stereo image pairs.

These features are used to clustering images by PVC and to train a BP

neural network in order to obtain an adaptive matcher. The trained BP

matcher can generate an initial or primitive disparity map that

provides necessary correlation or corresponding SSD (sum of squared

differences) in area-based matching methods. Following the BP

training, we propose a matching algorithm based on image clustering

that is useful in iteratively updating the primitive disparity map.

We show that this update algorithm can improve the quality of the

disparity map significantly. At the final stage, constraints such as

epipolar line, ordering, geometric and local-support are added to

refine the map. Empirical results show that the efficiency of the BP

matcher and the validity of our matching algorithm.

 

[科資中心編號]RB8707-0459

[題    名]

期刊自動登錄及目次文字辨識系統---子計畫I:期刊封面彩色影像圖文分離

Text Area Extraction from the Color Image of Journal Cover Pages

[研 究 者 ]

陳淑媛

Chen, Shu-Yuan

[機構名稱  ]元智工學院資訊工程系(YMGACES)

[經費來源  ]行政院國家科學委員會(NSC)

[補助金額  ]351700

[補助編號  ]NSC85-2213-E155-021

[研究開始日期]1995/08

[研究結束日期]1996/07

[頁 冊 數 ]86頁

[語    文]英文

[建檔日期  ]1999/10

[出版情形  ]完成日期:民國85年7月

[原件提供  ]行政院國家科學委員會科學技術資料中心

NSC85-2213-E155-021

[科資分類號 ]IA0403

[關 鍵 詞 ]

影像處理;圖文分離;彩色影像分割

 

Image processing;Text area extraction;Color image segmentation

 

[摘    要]

      區塊分割是文件分析過程中,極重要且前置的一個步驟。本篇報告提出一個從多樣型式的彩色文件中,特別是圖文互疊的複雜彩色文件,擷取出文字區塊的新系統。首先為了要降低全彩影像處理時的複雜度,我們提出了一個簡單的色彩量化方法,可以將24位元的全彩影像,量化成42個或更少的顏色群集。接著利用塗黑以及標記技術,配合邊緣及顏色資訊,找出各個性質相似的區塊。然後利用一些傳統及我們自己提出的文字特徵,判定區塊內是否為文字。最後,運用後處理程序來校正區塊屬性分類上的錯誤。另外,我們也進行複雜度分析,彈性的選用不同的處理步驟,使得用色較少、圖文不互疊的簡單文件,可以被視作單色影像,因而加快處理時間。而對於用色較多且圖文互疊的複雜文件,系統將花較長的處理時,但最終仍能得到正確的結果。當然,處理複雜文件所要付出的代價,便是處理時間的加長。實驗結果可以證實所提方法是可行且實用的。

 

      Page segmentation is a prior and primary process in the

document analysis task. In this report, a new adaptive page

segmentation system is proposed to extract text blocks from various

kinds of color document images, including those with texts overlaid

on graphics/images. First, in order to degrade the processing

complexity on true color images, a new simple quantization method is

employed to reduce the color numbers from 24-bit true colors to 42

clusters or less. Then, using smearing and labeling techniques

together with edge and color information, each coherent blocks can be

found out. After that, some conventional and new features are

computed from each block to decide whether it is a text block or not.

Finally, postprocessing is adopted to rectify the classification

results. In addition, complexity analysis is performed to adjust

processing steps such that the simple images, with few colors and

none texts-graphics/images overlapping, can be treated as monochrome

images to speed up processing time. While for the complex images,

with many colors and texts-graphics/images overlapping, more

processing time is required to obtain correct segmentation results.

That is the price paid for the complexity. Experimental results prove

the feasibility and practicality of the proposed approach.

 

[科資中心編號]RB8705-0147

[題    名]

電腦視覺技術與應用---子計畫二:以數學形態學從事距離影像分割及三維不變特

徵抽取之研究

Research on Range Image Segmentation and 3D Invariant Feature

Extraction by Using Mathematical Morphology

[研 究 者 ]

洪一平

Hung, Yi-Ping

[機構名稱  ]中央研究院資訊科學研究所(ASII)

[經費來源  ]行政院國家科學委員會(NSC)

[補助金額  ]364300

[補助編號  ]NSC85-2221-E001-001

[研究開始日期]1995/08

[研究結束日期]1996/07

[頁 冊 數 ]28頁

[語    文]英文

[出版情形  ]完成日期:民國85年10月

[原件提供  ]行政院國家科學委員會科學技術資料中心

NSC85-2221-E001-001

[科資分類號 ]IA0406

[關 鍵 詞 ]

距離影像分割;電腦視覺;特徵抽取;不變性特徵;數學形態學;影像處理

 

Range image segmentation;Computer vision;Feature extraction;Invariant

feature;Mathematical morphology;Image processing

 

[摘    要]

      在本計畫中,我們提出了一個利用數學形態學來抽取距離影像(Range

image)中的不變性特徵(Invariant feature)的方法。此處的「不變性」指的是

在三度空間的剛性變換(即旋轉加上平移)下,所抽取的特徵仍能維持不變。傳統上,距離影像的特徵抽取經常是利用微分幾何學裡提供的工具,例如法向量(Surface normal)或曲率(Curvature)來進行的。這些方法通常需要計算距離影像的一階甚至二階的微分。因此,這些方法極容易受到雜訊的影響。反之,數學形態學中的運算元先天上即具有可以濾除影像雜訊的特性,因此極適合被應用於距離影像的特徵抽取。傳統上,利用數學形態學從事影像特徵抽取的方法,通常只能抽取出距離影像的高度變化區或者是邊線特徵(Edge feature)。在本論文之中,我們提出了一個新的方法,可以在距離影像的高度變化區之中再區分出角(Corner)特徵以及邊線特徵。

      在此方法中所使用的是球狀的結構元以及Opening或Closing等數學形態學之運算元。首先,我們抽取出距離影像中的「高度變化區」。接著,我們發展出維度判定法,將高度變化區再區分出角特徵以及邊線特徵。這個方法在使用不同大小的結構元的方式下,能夠在多尺度(Multiscale)的架構中運作。而且,這個方法同時適用於幾何形狀及曲面形狀的物體。經由這個方法所得到的結果通常都十分符合人類直覺上對於角與邊線的認定。而且它們也都是三維剛性變換下的不變性特徵。本計畫的成果可應用於較高階的視覺處理工作,例如,不同視角觀測物面之定位及物件辨認。

 

      Mathematical morphology is inherently suitable for range image

processing because it can deal with the shape of a function in a

natural and intuitive way. In this work, a new approach to the

extraction of the corner-edge-surface structure from 3D range images

is proposed. Morphological operations are utilized for segmenting

range images into smooth surface regions and high-variation surface

regions, where the high-variation surface region are further

segmented into regions of edge type and regions of corner type. A new

3D feature, HV-skeleton, can be extracted for each high-variation

surface region. The HV-skeletons can be thought of as the skeletons

of high-variation surface regions and are useful for feature matching.

The 3D feature extracted by our approach are invariant to 3D

translations and rotations, and can be utilized for higher-level

vision tasks such as registration and recognition. Experimental

results show that the new 3D feature extraction method works well for

both simple geometric objects and complex shaped objects such as

human faces.

 

[科資中心編號]RA8401-2017

[題    名]

以馬可夫隨機場及神經網路基礎之影像處理系 統

Markov Random Field and Neural Network for Image Processing.

[研 究 者 ]

邱炳樟

CHIEU,BIN-CHANG

[機構名稱  ]台灣工業技術學院電子工程技術系 (NTITENT)

[經費來源  ]行政院國家科學委員會(NSC)

[補助金額  ]483000

[補助編號  ]NSC82-0404-E011-080

[研究開始日期]1992/08

[研究結束日期]1993/07

[出版日期  ]1994/07

[頁 冊 數 ]70

[語    文]中文

[建檔日期  ]1995/10

[科資分類號 ]IA;IA0403

[關 鍵 詞 ]

馬可夫隨機場;影像處理;神經網路;影像辨識; 影像分割

 

Markov random field;Image processing;Neural network;Image recognition;Image

segmentation

 

[摘    要]

      在馬可夫隨機場模式理論下,本文提出一 個影像辨識系

統。這個系統採用監督式的架構 而概分為兩階段來完成:訓

練階段與分類階段 。在訓練階段中,以梯度上昇方法作為重

複估 測吉伯斯參數向量的依據,最後以一具有最大 希望值的

最佳吉伯斯參數向量來代表某一類影 像,依此,每一類影像各

對應一吉伯斯參數向量 。在分類階段中,則從已知的各吉伯

斯參數向 量,選擇具有最大希望值的吉伯斯參數向量作 為某

一測試影像的分類。由於在訓練與分類階 段中,均牽涉到吉

伯斯分布函數的計算,作者乃 提出一個部分連接的格子結構

來完成所需的計 算。這格子結構,適合平行計算。再者,由於

兩 階段中均採用了最大希望法則,本系統總是具 有良好的行

為。基本上,本方法從共同機率著 手,解決了馬可夫隨機場的

兩個基本問題:吉伯 斯函數計算與吉伯斯參數向量估測,而其

應用 也不限於辨識問題或影像資料。

 

      In this thesis a supervised image recognition system based on Markov

random fields (MRF's) is proposed. The procedure creating it can be

approximately divided into two basic phases: the training phase and the

classification phase. In the training phase, the gradient ascent method is

employed for iterative estimation of the optimal Gibbs parameter vector

yielding maximum likelihood of a set of training images, and this optimal

parameter vector is then used to represent the class of the image data.

Accordingly, each known class of training images corresponds to a Gibbs

parameter vector. In the classification phase, a testing image is assigned to

the class whose corresponding Gibbs parameter vector provides the maximum

likelihood of the image among all known classes. Since both these two phases

involve computation of Gibbs distributions (GD's), a partially connected

lattice structure is developed to compute efficiently these GD's. This lattice

structure is suitable for fast parallel computing. In addition, due to use of

the criterion of maximum likelihood in the two phases, the system always

performs well. Basically, by the joint probability, this approach solves two

basic problems in MRF modeling: how to measure a GD and how to estimate the

Gibbs parameters from MRF samples. Its applications may not be limited to the

recognition problems nor the image data.

 

[科資中心編號]RA8304-2311

[題    名]

影像分塊方法及其界定值選定之研究

Methods of Image Segmentation and the Selection of Threshold.

[研 究 者 ]

羅慶昌

LO,KING-CHANG

[機構名稱  ]成功大學航空測量研究所(NCKUPHOG)

[經費來源  ]行政院國家科學委員會(NSC)

[補助金額  ]237000

[補助編號  ]NSC82-0410-E006-190

[研究開始日期]1992/08

[研究結束日期]1993/07

[出版日期  ]1993/10

[頁 冊 數 ]99

[語    文]中文

[建檔日期  ]1994/12

[科資分類號 ]IG;IG0001

[關 鍵 詞 ]

影像分塊;區塊匹配;界定值選定; 數值地形模型

 

Image segmentation;Region matching;Threshold selection; Digital elevation

model (DEM)

 

[摘    要]

      利用數值立體像對,經影像匹配方法來自 動產生數值地

形模型(DEM)為經濟而便捷之方式 。但影像匹配時,若在灰值

均調區內或相同紋 理區內實施,會造成匹配失敗。為解決此

問題, 乃選用灰值均調區及相同紋理區之邊界線實施 區塊匹

配。在實施區塊匹配前,必須先進行數 值影像分塊以產生區

塊匹配所需之邊界線,本 研究選用混合連結區塊成長法來對

灰值均調區 作分塊,及選用紋理頻譜分析法來對相同紋理 區

作分塊,成效良好。

      當在實施分塊時,其間經常需選定適當界 定值以供利用,

以往都使用試誤法,本研究使用 統計方法自動選定適當界定

值分別進行影像二 元化、邊緣線擷取與區塊成長等影像分

塊處理 。最後之區塊成長分塊法,結合了邊緣線法與 區塊法

的特性,特別是其中有關界定值部分,可 利用灰值直方圖統計

分析法算得灰值界定值、 相關係數法算得灰值梯度界定值,

不需人工干 預,可達到自動化的目的。

 

      Automatic digital elevation model (DEM) generation by image matching is

a economical and fast approach. But the matching would fail in the homogeneous

grey value regions or uniform texture regions. To solve this problem, we can

use region matching instead of intensity-base matching. Before region

matching, the image segmentation is needed. We select hybrid linkage region

growing method for segmentation of homogeneous grey value regions, and texture spectrum analysis method for segmentation of uniform texture regions. The performance is good in practice.

      During the segmentation procedure, we need to automatically select

adaptive threshold by statistical method for image binarization, boundary

detection and region growing. The method of gray value histogram statistical

analysis is used for automatical gray value thresholding, and correlation

criterion method is used for thresholding the gradient of gray value. Then,

the method which integrates edge-based method and regional growing method is

performed for segmentation. The performance is evaluated by using the

criterion functions such as uniformity and shape measures.

 

[科資中心編號]RA8304-2017

[題    名]

以解析方式而得的分群技巧及其應用

Analytical Approach of Clustering Techniques with Applications.

[研 究 者 ]

林志青

LIN,JA-CHEN

[機構名稱  ]交通大學資訊科學研究所(NCTUINSG)

[經費來源  ]行政院國家科學委員會(NSC)

[補助金額  ]320000

[補助編號  ]NSC81-0408-E009-589

[研究開始日期]1992/02

[研究結束日期]1993/01

[出版日期  ]1993/05

[頁 冊 數 ]67

[語    文]英文

[建檔日期  ]1994/12

[科資分類號 ]IA;IA0401;IA0403

[關 鍵 詞 ]

分群;平行處理;影像壓縮;影像切割; 解析式兩群分群法;多群

分群法;分界線

 

Clustering;Parallel processing;Image compression;Image segmentation;

Analytical two-class clustering technique; Multiple-class clustering

technique;Decision boundary

 

[摘    要]

      本報告提出一種新的分群法,包括多群分 群法及解析式

兩群分群法。實驗結果顯示解析 式兩群分群法可以將所要

兩群的群中心及分界 線很快地求出。當資料不見得是兩群

,我們 仍可使用此兩群分群技巧,其方式如下述。首 先我

們將原始資料用兩群分群技巧將其一分為 二,然後在兩個子

集合裡,再各自使用兩群分群 技巧而使其二分為四,重複這步

驟直到每個子 集合滿足本報告所提的群中心密度判別式,且

每條分界線滿足本報告所提的分界線密度判別 式。最後,再

將那些緊緊相鄰的子集合合併即 可。實驗結果及平行處理

的討論均包括在本報 告內。本分群法的三度空間版本可直

接用在影 像處理,諸如壓縮及切割等等。

 

      A new method for multiple-class clustering is proposed in this report.

This method is based on the analytical two-class clustering technique which is

also derived in this report. The analytical two- class clustering technique is

observed to be able to cluster a two- class data set and obtain quickly the

cluster representatives of the two classes and the decision boundary to

seperate them. The analytical two-class clustering technique is then applied

to cut a given input set, which is not necessary to be formed of two clusters,

into two subsets. For each of these two subsets, two-class clustering

technique is applied again so that four smaller subsets are generated after

cutting. This procedure is repeated again and again until every subset meets a

proposed Mean-density criterion and every decision boundary generated by

two-class clustering technique meets a proposed Line-density criterion. The

output of the above cutting procedure are several subsets whose union is the

input data set. These subsets are then merged according to the law that any

two subsets whose centroids are close to each other should be merged into one

unless the regions expanded by these two subsets do not overlap. Experimental

results are given, and the way to implement the proposed method in parallel is

also discussed. The 3D version of the proposed clustering method can be

applied in image processing such as compression or segmentation.

 

[科資中心編號]RB81-1115

[題    名]

概率式核醫影像分割法

Probabilistic Radionuclide Image Segmentation.

[研 究 者 ]

吳良治;劉仁賢

WU,LIANG-CHIH;LIU,REN-SHYAN

[機構名稱  ]榮民總醫院核子醫學部(VGHGS)

[經費來源  ]行政院國家科學委員會(NSC)

[補助金額  ]589200

[補助編號  ]NSC81-0412-B075-47

[研究開始日期]1991/08

[研究結束日期]1992/07

[出版日期  ]1992/07

[語    文]中文

[建檔日期  ]1994/05

[科資分類號 ]EA02;EA0204

[關 鍵 詞 ]

影像分割;概率式

 

Image segmentation;Probabilistic

 

[摘    要]

      影像分割技巧通常利用單一影像局部特性 來處理,以使

器官與其背景分離。本計畫則擬 由器官功能特性,以概率方

式來分割動態系列 之不同功能部分。

      以肝臟檢查為例,根據肝臟動態檢查的60張 圖像(每張一

分鐘),我們可求出兩種參數:稱為 BUT(Built up time)和

PUR(Parenchymal uptake ratio)。其中, BUT定義為每個像素(Pixel)活性

達到其極大值之 0.693倍所花之時間,而PUR則定為每一像素在5

分 鐘時之活性與其活性極大值之比,在建立資料 庫時,我們

先以人為方式選定Blood pool、Liver parenchyma、Hepatic duct和

Intestine四大部分。因此, 我們有四組二維樣本空間,其概率密

度函數 (Probability density function)可以二維常態分布函數 推定

。在對任何病人資料作影像分割時,為去 除背景因素,我們先

BUT影像作Histogram分析,在 BUT的Histogram中,第二組Peak附近所

對應之像素通 常為肝細胞(第一組為Blood pool),接著我們將對

BUT第二組Peak的極大值圖像(此圖像定義為各像 素在此動態系

列中之極大值)求取平均值,此值 25%以下的區域設為背景,這

樣我們就可以去除 背景因素。對任一像素(含有BUT與PUR兩參

),由 前述已知取樣之概率密度函數和相同之事前機 率,我

們可使用Bayes' theorem求得每一像素歸屬於 上述四種組織之概

,取其最大值,即可得到一 張分割完成之影像。同理,此方

法亦可用於其 他檢查。

      研究的結果顯示,概率式的影像分割法可 自動分割動態

系列之不同功能部分。

 

      Image segmentation is usually done in spatial domain using local

characteristics of the pixels in the image. This study was undertaken to

segment dynamic radionuclide studies according to their functional properties

in a probabilistic way.

      Choosing hepatobiliary study as an example, dynamic images were acquired

in 60 frames for 60 minutes. Two parameters, namely, built-up time (BUT) and

parenchymal uptake ratio (PUR), were derived from the series of images. BUT

was defined as the time required for each pixel to reach 0.693 of its maximum

activity. PUR was defined as the ratio of the activity at 5 minute to the

maximum activity of the corresponding pixel. Four groups of samples were

manually drawn from the blood pool, liver parenchyma, hepatic duct, and

intestine. To get rid of the background region, histogram analysis was done on

the BUT image. BUTs around the second peak of the histogram were regarded as

the time for liver parenchyma. Pixels with activities less than 25% of the

average activities of pixels at the second BUT peak were masked out. After the

estimation of probability density functions of the four groups from the

samples, the probability of each pixel being in the four groups was calculated

with equal a priori probabilities. Image could then be segmented according to

the highest probability of each pixel in the four groups. Similar approach can

be done on others such as cardiac study.

      In summary, the results of the study demonstrates that dynamic

radionuclide images can be automatically segmented into different functional

parts in a probabilistic way.

 

[科資中心編號]RA81-2181

[題    名]

彩色影像分割

Color Image Segmentation.

[研 究 者 ]

曾定章

TSENG,DIN-CHANG

[機構名稱  ]中央大學電機工程系(NCUEELE)

[經費來源  ]行政院國家科學委員會(NSC)

[補助金額  ]441000

[補助編號  ]NSC80-0408-E008-11

[研究開始日期]1991/02

[研究結束日期]1992/01

[出版日期  ]1992/01

[頁 冊 數 ]60

[語    文]英文

[建檔日期  ]1994/05

[科資分類號 ]IA;IA0403

[關 鍵 詞 ]

彩色影像;影像分割;IHS彩色模式; 色彩均等分布;群聚法

 

Color image;Image segmentation;IHS color model; Uniform chromaticity

scale;Clustering

 

[摘    要]

      影像分割是電腦視覺及圖形識別研究上一 項極為基本

且重要的工作。在這個研究計畫中 我們提出了一個以色彩

為基礎的影像分割方法 來分割彩色影像。

      首先我們就目前已有的色彩空間(Color space)做一分析與

比較。而後挑選出一個具有三 維色彩均等分布(3-D uniform

chromaticity scale)且與 人類視覺系統(Human visual system)相吻合的

色彩空 間來將其轉變成IHS的型態。然後分析IHS色彩空 間的

各個屬性(Attribute)的特性及其相關性來做 為我們提出分割方

法的依據。

      我們的分割方法包含了幾個觀念,一是訂 定色彩空間的

有色(Chromatic)及無色(Achromatic)區 域範圍,二是色彩空間各屬

性的梯圖分割 (Histogram thresholding),三是過度分割

(Oversegmentation)的處理方式。依各種不同應用的 需求,我們可

以從上述的觀念推演出各種不同 的處理方式(Approach)。例如

我們可以利用上述 的觀念推演出擷取(Extract)彩色影像中某

種特定 顏色區域的處理方式。這種處理方式可應用於 無人

自動車(Autonomous land vehicle)的找路問題上, 可去除影像中陰影

(Shade)的影響因素,可分離墨 水汙染支票上的字跡,可擷取彩

色文件、地圖 上不同顏色的資訊...等等。我們也可利用上

述 的觀念來推演出分割彩色影像成一些有意義區 域的處理

方式。例如一朵花的影像不會因花瓣 的新舊而被分割成好

幾個區域,或一條路的影 像不會因路上樹木、建築物的陰影

而被分割成 許多區域...等。在我們的計畫中包含上述多種

不同應用的處理方式。

 

      Image segmentation is an essential and important task for computer

vision and pattern recognition. In this project a general approach for

achieving color image segmentation using uniform-chromaticity-scale perceptual

color attributes is proposed. At first chromatic and achromatic areas in a

perceptual IHS color space are defined. Then image is separated into chromatic

and achromatic regions according to the region locations in the color space.

1-D histogram thresholding for each color attribute is performed to split the

chromatic and achromatic regions. At last region growing is used to solve the

oversegmentation problem. In experiment we demonstrate the power of the

proposed approach. We derive algorithms from the general approach to segment

flower image, to segment road image for autonomous land vehicle, to extract

shade-road in an image, to extract color-based information in color documents,

etc. 

 

科資中心編號: RB9104-0068 

題名: 資訊檢索中索引典系統建置之研究

A Study on Construction of Thesaurus Systems for Information Retrieval 

研究者: 陳光華

Chen, Kaung-Hwa 

機構名稱: 台灣大學圖書及資訊管理系 

經費來源: 行政院國家科學委員會 

補助金額: 新台幣 456 仟元 

補助編號: NSC89-2213-E002-112 

研究日期: 2000/08 - 2001/07 

出版日期: 2001/07 

頁冊數: 6頁 

原件提供: 行政院國家科學委員會科學技術資料中心, NSC89-2213-E002-112 

科資分類號: IA0407;IA0304;IA0410 

關鍵詞: 資訊檢索 ;索引典系統 ;數位圖書館 ;控制詞彙 ;檢索點

Information Retrieval ;Thesaurus System ;Digital Library ;Controlled Vocabulary ;Access Point 

中文摘要: 從WWW 上資訊檢索應用而言,資源必須經過有系統地組織與整理,才能得到滿意的檢索結果。控制詞彙的角色在前述的背景下,扮演越來越關鍵的角色。控制詞彙通常是透過索引典或是主題表體現,提供詞彙之間的等同關係、層級關係、關連關係。藉由索引典,使用者可以瞭解資訊系統的知識結構,可以選擇適切的檢索詞彙,以獲致比較滿意的檢索結果。然而,索引典的建構與管理卻是一項困難工作,若沒有設計良好的系統,則無法有效地發揮索引典的長處。目前也沒有任何的資訊檢索系統或是數位圖書館系統整合索引典子系統,使得資訊檢索系統的效能始終侷限於求準率(Precision)不高的窘況。本計畫探討索引典的建置,發展索引典系統的功能模式,建立索引典建置與管理之模式,並實作雛型系統。 

英文摘要: From the viewpoint of WWW-based IR researches, the metadata and resource description framework become much more important than ever before. Therefore, the application of the controlled vocabularies increasingly plays an important role. The recent literatures also report that controlled vocabularies significantly improve IR effectiveness. Thesauri and Subject Headings consist of controlled vocabularies and represent the knowledge structures in details and in various relations. However, the construction and management of thesauri is not an easy task and the integration of thesaurus subsystem into IR systems is not considered in the current IR systems or DL projects. The project will investigate the applications of thesauri and subject heading, propose a model to handle the thesauri or the subject headings in a systematic way, and implement a practical system. 

 

 

科資中心編號: RB9102-0970 

題名: 以分散式物件技術整合異質性資料檢索技術

Integrating Heterogeneous Information Retrieval Technology through Distributed Object-Oriented Paradigm 

研究者: 張玉山 ; 袁賢銘

Chang, Yue-Shan 

機構名稱: 明新技術學院電子工程系 

經費來源: 行政院國家科學委員會 

補助金額: 新台幣 333 仟元 

補助編號: NSC89-2626-E159-001 

研究日期: 2000/08 - 2001/07 

出版日期: 2001/07 

頁冊數: 6頁 

原件提供: 行政院國家科學委員會科學技術資料中心, NSC89-2626-E159-001 

科資分類號: IA0202;IA0301 

關鍵詞: 資訊檢索 ;系統整合 ;共通物件策託經紀人架構 ;分散式物件 ;可擴展標示語言

Information Retrieval ;System Integration ;Common Object Request Broker Architecture (Corba) ;Distributed Object ;Extensible Markup Language (Xml) 

中文摘要: 在這個計畫中我們以CORBA 架構為基礎設計一個可以整合各種不同資訊檢索系統的存取介面,這個介面可以整合各種不同的資訊源,為了達成整合各種不同的資訊源,系統中以XML 的DTD 來描述各種資訊源的中介資料(Metadata),架構在我們所設計的架構下我們也設計一個異質性資訊檢索的平台,這個平台整合了一般的搜尋引擎、論文檢索及擷取系統及圖書資料檢索及擷取系統。 

英文摘要: In this project, we propose an Integrated Information Retrieval (IIR) service based on the Common Object Request Broker Architecture (CORBA) and apply the Document Type Definition (DTD) of eXtensible Markup Language (XML) to define the metadata of information sources The objective of using the IIR design is not only to provide programmers with a uniform interface for coding a software application that can query a variety of information sources on the Internet, but also to create a flexible and extensible environment that easily allows system developers to add new or updated wrappers to the system. 

 

 

科資中心編號: RB9102-0304 

題名: 子計畫(III)無線通訊環境下以語音存取的互動式資訊檢索技術

Interactive Information Retrieval Techniques with Spoken Access Capability for Wireless Communications Applications 

研究者: 簡立峰

Chien, Lee-Feng 

機構名稱: 中央研究院資訊科學研究所 

經費來源: 行政院國家科學委員會 

補助金額: 新台幣 483 仟元 

補助編號: NSC89-2213-E001-051 

研究日期: 2000/08 - 2001/07 

出版日期: 2001/07 

頁冊數: 4頁 

原件提供: 行政院國家科學委員會科學技術資料中心, NSC89-2213-E001-051 

科資分類號: IA0304;IA0400 

關鍵詞: 動態索引典 ;資訊檢索 ;語音互動 ;無線通訊

Dynamic Thesaurus ;Information Retrieval ;Speech Interaction ;Wireless Communication 

中文摘要: 本計畫的目的是發展以動態索引典(Dynamic Thesaurus) 為基礎的互動式檢索技術(Interactive Retrieval)以提高網路資訊檢索系統的互動性及正確率,另外發展適合無線通訊環境利用的語音輸入查詢技術,希望藉以擴大無線通訊環境的資訊利用空間。本計畫的研究為期三年,計畫所欲發展的技術方向將包括動態索引典(Live Thesaurus) ,強健語音檢索 (RobustSpeech Retrieval),語音互動技術 (SpeechInteraction)等。這些技術將有效整合其他子計畫的技術發展。本計劃第一年研究已經產生成果包括會議論文7 篇其中較為重要包括log-based term suggestion,interactive search, anchor text mining forterm translation extraction 等。 

英文摘要: The purpose of this project is intended to develop interactive retrieval techniques with spoken access capability for the enhancement of retrieval effectiveness in a wireless communication environment . The first mission of this project is to allow users can retrieve network information via speech communication. To do so, successful integration with speech recognition and information retrieval technologies is very important. Another mission of the project is to deal with so-called short query problems, by means of the development of live thesaurus construction and term suggestion techniques . The ongoing is a three-year project. During the three years, the research topics to be explored will include live thesaurus construction, robust speech retrieval and speech interaction techniques. In the first year, some preliminary results have been obtained. There have been published 5 papers and several related techniques developed such as log-based term suggestion, interactive search, anchor text mining for term translation extraction, etc. 

 

 

科資中心編號: RB9101-1015 

題名: 以低成本調適知識學習法解析非限制性文本詞義之研究

A Study on Low-Cost Adaptive Knowledge Learning Scheme for Disambiguating Word Sense in Unrestricted Tesxts 

研究者: 陳振南

Chen, Jen-Nan 

機構名稱: 銘傳大學資訊管理系 

經費來源: 行政院國家科學委員會 

補助金額: 新台幣 336 仟元 

補助編號: NSC89-2213-E130-010 

研究日期: 2000/08 - 2001/07 

出版日期: 2001/10 

頁冊數: 5頁 

原件提供: 行政院國家科學委員會科學技術資料中心, NSC89-2213-E130-010 

科資分類號: IA0407 

關鍵詞: 語意網路 ;知識擷取 ;調適性學習 ;統計式詞彙語意分析 ;索引典

Semantic Network ;Knowledge Extraction ;Adaptive Learning ;Statistical Word Sense Disambiguation ;Thesaurus 

中文摘要: 一詞多義是語言的普遍現象,語意解析就是要依靠詞彙的前言後語及句子的上下文處理一詞多義之問題。詞彙語意解析常被公認為是自然語言處理應用系統中的首要工作。近年來詞彙語意解析一直成為計算語言學界非常熱門的議題,尤其是語意的處理可應用在各種不同的自然語言相關處理中,如大家所熟悉的機器翻譯、資訊檢索、文件摘要、文件分類、資訊擷取等。雖然自動化之語意解析一直受到關注,然而其所面臨之瓶頸在於特定領域之知識取得不易。本研究將建構一個自動調適語意知識的系統,以改進過去作法之缺點。 

英文摘要: Word sense disambiguation (WSD) is the problem of assigning a sense to an ambiguous word by using its context. The goal of WSD is expected to improve the analytical accuracy for tasks that are sensitive to lexical semantics. Such tasks include machine translation, information retrieval, text summarization and so on. Recently, various methods are proposed to resolve this task. However, data sparseness is an inherent problem in previous methods for natural language processing. In this work, we propose an adaptive scheme for disambiguating word sense in unrestricted texts using available information in machine-readable dictionary. An evaluation of the method was done on both the Wall Street Journal(WSJ) and on Brown Corpus. The percentages of correct resolution were achieved with this two corpora were WSJ 77.3% and Brown 70.5%. Such results were suggested that our adaptive approach is quite a stable one. 

 

 

科資中心編號: RB9101-0278 

題名: 詞彙為本的知識連結---朝向多語數位圖書館中之詞彙網路基礎架構

Lexicon-Based Knowledge Linking---Approaches towards a WordNet Infrastructure for a Multilingual Digital Library 

研究者: 黃居仁 ; 陳克健 ; 柯淑津

Huang, Chu-Ren ; Chen, Keh-Jiann ; Ker, Sur-Jin 

機構名稱: 中央研究院語言學研究所籌備處 

經費來源: 行政院國家科學委員會 

補助金額: 新台幣 1090 仟元 

補助編號: NSC89-2750-P001-011-2 

研究日期: 2000/08 - 2001/07 

出版日期: 2001/07 

頁冊數: 4頁 

原件提供: 行政院國家科學委員會科學技術資料中心, NSC89-2750-P001-011-2 

科資分類號: IA0405;IA0202 

關鍵詞: 詞彙為本知識連結 ;多語數位圖書館 ;數位圖書館 ;雙語詞彙網路 ;語際指引 ;本體論

Lexicon Based Knowledge Link (Lkl) ;Multilingual Digital Library ;Digital Library ;Bilingual Wordnet ;Interlingual Index (Ili) ;Ontology 

中文摘要: 國際數位圖書館必然需要多語言、跨語言的資訊處理能力,亦即擁有多語語文座標與知識管理的技術,方能建構理想的使用環境。一個理想的方案是建立多語的詞彙語意網路作為資訊處理的核心基礎。我們採行的具體作法是先建立一個中英雙語詞網,而後再利用現有的國際學術資源-歐語詞網,進一步形成多語的詞彙語意連結,以作為資訊處理技術的基礎。本計畫是建構中英雙語詞彙網路的第一步;目標將限定在中文詞彙與英語詞網節點的對應上。建構完成的對應資料庫將可進而轉換、修正與擴充為中英雙語詞網。 

英文摘要: A critical foundational issue in digital library is to find an ontology that is both universally accepted and efficient in describing knowledge content. It's the goal of this project to provide such ontology in the form of a bilingual wordnet. By constructing a linguistic ontology, we will take the first step towards a unified solution of linguistic anchor and knowledge management in the DL environment. In this project, we focus on English/Chinese bilingual wordnet. It will be the basis of constructing both a comprehensive ontology for Chinese Digital Library, as well as linking (through English and ILI), to wordnets, and hence ontologies, of other languages. In the first phase of the project, the ontology skeleton is built with a set of Lexicon-based Knowledge Links (LKL). LKL is conceived as a set of (non-hierarchical) semantic relations that provide both basic referential information for all crucial DL activities, such as multilingual knowledge extraction, information retrieval, document classifications etc. 

 

 

科資中心編號: RN9101-0320 

題名: 以使用者查詢記錄產生之控制詞彙表在網路資訊檢索之應用研究

User Controlled-Vocabulary from Web Query Terms for Web Information Retrieval 

研究者: 卜小蝶 

機構名稱: 世新大學圖書資訊系 

經費來源: 行政院國家科學委員會 

補助編號: NSC89-2413-H128-006 

研究日期: 2000/08 - 2001/07 

出版日期: 2001/10 

頁冊數: 7頁 

原件提供: 行政院國家科學委員會科學技術資料中心, NSC89-2413-H128-006 

科資分類號: PA0210 

關鍵詞: 資訊檢索 ;網際網路 ;檢索詞彙分析 ;自動分類 ;使用者查詢記錄 ;控制式詞彙

Information Retrieval ;Internet ;Search Term Log Analysis ;Automatic Classification ;User Search Log ;Controlled Vocabulary 

中文摘要: 近來國際上越來越重視以檢索詞彙記錄來分析使用者的資訊行為,加以本計劃主持人先前研究發現台灣地區使用者具有檢索詞彙簡短、存在核心詞彙、及檢索專有名詞比例高等特性,並初步瞭解高頻詞彙的主題類別分佈情形。本研究則進一步提出一整合人工分析及電腦自動的主題分類方法,將檢索詞彙以事先製定之主題類別進行分類。研究設計包括主題類別的製定、特徵詞彙的抽取、特徵詞彙的人工分類、及未知類別檢索詞彙的自動分類等步驟。實驗資料包括在不同時期所收集的二種搜尋引擎檢索詞彙記錄,並以熱門主題類別為範疇,將人工分類與所提之自動分類方法結果進行比較。實驗結果顯示,本研究所提方法能有效地將大量檢索詞彙進行分類,且所建議之主題類別與人工分類結果相近。本研究成果為建立特定領域控制詞彙表的基礎工作,因此,對於改善網路搜尋系統的檢索效益、與瞭解網路使用者的資訊需求等皆具有相當的應用價值。 

英文摘要: There is an increasing research interest in investigating Web users’ information behavior by using search engine logs. Continuing the author’s previous research, this study presents a search term categorization approach to automatically classifying Web search terms into broad subject categories. Two search engine logs in Taiwan were collected and tested from different periods of time. The achieved performance is quite encouraging compared with that of human categorization. Through good integration of human and machine efforts, the approach can be taken as the basis for constructing user-oriented controlled vocabulary. This has shown potential for use in various applications like enhancing Web information retrieval effectiveness and observing users’ search interests and. 

 

 

科資中心編號: RF9101-0376 

題名: 以語意類似程度為基礎之數位化圖書館資訊查詢方法

A Semantic Similarity Degree Approach to Information Retrieval for Digital Library 

研究者: 黃明祥

Huang, Ming-Hsiang 

機構名稱: 屏東科技大學資訊管理技術系 

經費來源: 行政院國家科學委員會 

補助金額: 新台幣 369 仟元 

補助編號: NSC89-2416-H020-013 

研究日期: 2000/08 - 2001/07 

出版日期: 2001/07 

頁冊數: 8頁 

原件提供: 行政院國家科學委員會科學技術資料中心, NSC89-2416-H020-013 

科資分類號: PA0210 

關鍵詞: 數位化圖書館 ;資訊檢索 ;語意近似 ;語意類似程度 ;資料表示

Digital Library ;Information Retrieval ;Linguistic Approximation ;Semantic Similarity Degree (Ssd) ;Data Representation 

中文摘要: 近年來,數位化圖書館(Digital Library, DL)的應用日益普及,不論是在學術界或商業方面的應用價值不斷的增加,數位化圖書館逐漸形成資訊管理一個重要的研究領域。對數位化圖書館的應用而言,查訊是一個相當重要的功能,由於目前使用的資訊查詢方法並不能有效解決數位化圖書館資訊查詢的需求,例如語意索引、結構交換、物件辨認與異質性資料整合等問題,主要的原因是處理語意方面的能力不足。有鑑於此,本研究提出一種以語意類似程度為基礎的方法,解決上述數位化圖書館資訊查詢方面的問題,本研究提出的方法主要優點在於具有豐富的語意處理能力以及快速資訊查詢的功能,對於數位化圖書館資訊查詢的功能具有極大的助益。為驗證本研究提出方法的可行性,本研究發展一套小型的數位化圖書館的雛型系統,測試語意查詢的功能與績效,作為發展大型數位化圖書館的基礎。本研究發現在特定領域的範圍,語意資料查詢的方法確實可以大幅度改進數位化圖書館的查詢能力,本研究亦提出一些具有發展潛力的研究方向,供後續研究者之參考。總而言之,本研究主要的貢獻如下:(1) 提出一個計算數位化圖書館資料之間的類似性之語意類似程度演算法(algorithm)並予以應用在數位化圖書館的資訊查詢作業,可以解決語意處理方面的問題,(2)發展一個應用於數位化圖書館之資訊查詢方法,改進目前查詢作業的效率,(3)發展完成一個應用於數位化圖書館之資訊查詢工具,以提升查詢工作的效率與品質。 

英文摘要: Recently, research on digital library has become an important research area due to its significant contributions to the practices and academics. Information retrieval (IR) plays an important role in the applications of DL. However, the ability of IR on DL is not satisfactory so far, especially in the aspect of semantic inquiry. The purpose of this research is to improve the ability of IR on DL, a method of semantic IR (SIR) based on Semantic Similarity Degree (SSD) is proposed. A small scale DL is developed to illustrate the application of the new method. Our results show that the ability of IR can be significantly improved in the process of IR on DL. The advantages of our approach and research issues are also discussed. Finally, some promising research topics are pointed out for further research. In summary, the major contribution by this research includes: (1) an algorithm based on SSD for IR is proposed, (2) an efficient method of IR for DL is developed, and (3) an efficient tool of IR for DL is developed. 

 

 

科資中心編號: RB9101-1006 

題名: 快速可拓模糊模型之設計及其在資訊檢索系統上之應用

The Design of Fast Extension-Based Fuzzy Model and Its Application to Information Retrieval System 

研究者: 黃有評

Huang, Yo-Ping 

機構名稱: 大同工學院資訊工程系 

經費來源: 行政院國家科學委員會 

補助金額: 新台幣 441 仟元 

補助編號: NSC89-2213-E036-050 

研究日期: 2000/08 - 2001/07 

出版日期: 2001/07 

頁冊數: 6頁 

原件提供: 行政院國家科學委員會科學技術資料中心, NSC89-2213-E036-050 

科資分類號: IA0407 

關鍵詞: 可拓模糊模型 ;彙總式搜索引擎 ;資訊檢索 ;查詢關鍵字 ;網站

Extenics Based Fuzzy Model ;Meta-Search Engine ;Information Retrieval ;Query Keyword ;Website 

中文摘要: 我們將上期計畫所發展的快速可拓模糊推論模型,在本期計畫裡成功的應用在智慧型的資訊檢索中。我們設計了一套智慧型彙總式搜索引擎,他能幫助使用者從輸入關鍵字查詢各大入口網站回應的巨大文件裡,自動找出較相關的文件,依權重由大至小排序給使用者閱讀。除此之外,此搜尋引擎將搜尋結果放入交談式視窗,讓使用者調整各相關詞彙的權重,並將搜尋結果的適用性回饋給系統,以做為將來修正模型之用。本文將說明此系統的架構並舉例說明我們的設計理念與測試結果。 

英文摘要: A previous developed fast extenics-based fuzzy model is applied to an intelligent information retrieval system in this project. We design an intelligent meta-search engine that bases on user's query keyword(s) to find the more related websites. Instead of listing a lot of candidate websites waiting for users to search, our approach automatically pops up only the more related documents for the user to read. Besides, the more related documents appeared in the dialogue window can be readjusted by the proposed method. User's satisfaction of the search result gives a feedback to our system for further improvement of the designed model. We will present the architecture of designed model and illustrate how the proposed system works in finding the desired websites. 

 

 

科資中心編號: RB9012-0393  

題名: 大型資訊系統可擴充式系統架構之設計(I)

Design of Scalable System Architectures for Large-Scale Information Retrieval Systems (I) 

研究者: 陳添福

Chen, Tien-Fu 

機構名稱: 中正大學資訊工程系 

經費來源: 行政院國家科學委員會 

補助金額: 新台幣 606 仟元 

補助編號: NSC89-2218-E194-005 

研究日期: 2000/08 - 2001/07 

出版日期: 2001/07 

頁冊數: 5頁 

原件提供: 行政院國家科學委員會科學技術資料中心, NSC89-2218-E194-005 

科資分類號: IA0407;IA0301 

關鍵詞: 資訊檢索系統 ;索引 ;反向檔案 ;群集指標 ;延展性

Information Retrieval System ;Index ;Inverted File ;Posting Clustering ;Scalability 

中文摘要: 資訊檢索系統變得很大而且很慢,因為大量的資訊以指數方式成長。最主要的資料檢索系統的結構是記錄每一筆文件位址的索引。當文件的條件數目增加時,索引的大小也會增加。索引技術最常用的是反向檔案,其中對每一筆文件的條件都記錄一個。為了減少反向檔案的大小而不犧牲查詢的效率,我們提出群集文件指標方案。我們依據資訊檢索系統的重點收集指標列到群集中並對每一群集以二進制編碼或位元向量。 

英文摘要: The most used index technique is the inverted file, which records the documents contain the term for each term. To reduce the size of the inverted file without sacrificing the query efficiency, we propose the posting clustering schemes. We collect posting sequences into clusters and encode each cluster with binary code or bit vector according to the emphasis of the information retrieval system. 

 

 

科資中心編號:RE9010-0003

題名:子計畫三:網路中文資源自動擷取技術的研究(I)中文文件特徵抽取技術

     Automatic Information Extraction for Chinese Network Resources(I)

     Feature Extraction of Chinese Documents

研究者:簡立峰

       Chien, Lee-Feng

機構名稱:中央研究院資訊科學研究所

經費來源:行政院國家科學委員會

補助金額:新台幣 415 仟元

補助編號:NSC86-2745-E001-001

研究日期:199608-199707

出版日期:199709

頁冊數:58頁

語文:中文

原件提供:行政院國家科學委員會科學技術資料中心

科資分類號:IA0406;IA0202;IA0407

關鍵詞:資訊檢索;特徵抽取;中文網路資源;自然語言處理;關鍵詞抽取

       Information Retrieval;Feature Extraction;Chinese Network Resource;

       Natural Language Processing;Keyword Extraction

摘要:目前網路上中文資訊成長迅速,但適合檢索中文資訊的高效率資源檢索系統則相當缺乏,發展適合檢索中文資訊的高效率檢索系統在目前看來是相當迫切與刻不容緩。本子計畫預計以三年時間發展出高效率、具備資源自動發現、過濾、擷取技術、以語言知識處理為基礎、適合檢索網路中文資源的智慧型網路資源搜尋系統。本計畫過去一年的研究成果相當豐碩,包括基礎研究、技術發展、系統開發、推廣應用都有相當成果。計完成如下列學術論文11篇及將尋易(Csmart)資訊檢索系統發展成可自動收集網路中文資源並施行中文檢索的系統。其進一步說明如下:(1)Text Retrieval System Design:發展出適合中文的特徵檔(Signature File)索引技術(ICASSP'97,ROCLING' 96),並以此特徵檔為基礎,進一步發?i出高效率自然語言查詢與Ranking技術,其檢索自由度大、精確率極高,並且適合檢索大量資料,如能成功應用於卓越商情的百萬篇產業訊息簡報摘要檢索。( 2)Network Information Retrieval Techniques:完成簡易資源自動收集發現技術(Robots)(IROL'96),發展出初步資源發掘技術(Spider),可自動獲取網路資源如網路新聞與Web Pages;另外可以成功結合語音與語言處理技術,以期進一步發展智慧型檢索技?N。(3)Automatic Keyword Extraction:改進傳統關鍵詞自動抽取技術過度依賴辭典且不易判斷專有名詞的缺點,發展出無須複雜辭典協助可從文件中自動抽取關鍵詞特別是專有名詞技術,這項技術應用寬廣對中文處理有基礎貢獻(ICCPOL'97a,SIGIR'97)。(4)I nformation Retrieval for Natural Language Processing:開始嘗試何利用網路源源不斷的資源以及高效率檢索技術,同時提昇自然語言處理技術(ICASSP'97)。此研究方向相當創新,國際學術界已經開?l重視。(5)Applications:進行若干資訊服務與技術測試,包括96年5月將部份技術成功應用在教育部16萬詞多媒體國語辭典檢索系統,6月完成即時新聞資料庫上線服務,並以經濟部ITIS產業資料庫進行實驗。11月更進一步以卓越商情百萬篇資料作實驗,12月應用於輔大圖書館50萬本圖書書目檢索,97年5月完成2GB的BBS資料庫測試。

      With the fast growth of Internet resources, high-performance information

     retrieval techniques and systems are highly demanded. Considering the late

      development of Chinese information techniques for  the Internet, this

     project is intended to develop several fundamental techniques and create

     efficient IR system, based on the need of Chinese, within three years.

     After the first year, some  achievements have produced. Eleven papers have

      been published and an improved NIR system constructed. In addition,

     techniques including text retrieval system design, network information

     retrieval  techniques, automatic keyword extraction, information retrieval

      for natural language processing have been developed and extensively

     tested.

 

科資中心編號:RB8910-0318

題名:智慧型知識擷取技術與應用研究(III):語料庫之設計與製作(III)

     Design and Implementation of Corpora (III)

研究者:陳光華

       Chen, Kuang-Hua

機構名稱:台灣大學圖書及資訊管理系(NTULIS)

經費來源:行政院國家科學委員會(NSC)

補助金額:新台幣 459 仟元

補助編號:NSC88-2213-E002-035

研究日期:199808-199907

語文:中文

原件提供:行政院國家科學委員會科學技術資料中心 NSC88-2213-E002-035

科資分類號:IA0407;IA0304

關鍵詞:語料庫;資訊檢索;標竿;自然語言處理;相關判斷

       Corpus;Information Retrieval;Benchmarking;Natural Language Processing;

       Relevance Judgment

摘要:在國內資訊檢索研究已日趨受到重視,合適的測試評估機制卻十分缺乏的背景下,本研究實際進行測試集的規劃與建置工作。測試集建構工作主要包括蒐集整理文件、建立查詢主題、以及進行相關判斷三個部分。本研究建立的文件集來源為新聞網站中的五種電子報,共有132,207篇文件。查詢主題是透過網路問卷實際徵集查詢需求,並進行三次的篩選之後,修正建構而成,共完成50個查詢主題。相關判斷的部分則是先對每個查詢主題建立一相關文件候選集,再針對候選集中的每篇文件以人工進行相關判斷,每一查詢主題由三位次判斷者同時進行,最後,則依據判斷結果計算並定義文件的相關程度。經由研究結果的分析顯示,本測試集有完整的架構及一定的規模,未來的研究應可以此為基礎,作進一步的擴展與改進。

      The research and development of information retrieval (IR) has made much

     progress recently. However, there's not any applicable mechanism for

     system evaluation in the Chinese research society. This  project aims at

     the design and the implementation for Chinese information retrieval

     benchmark. Generally speaking, a benchmark consists of a set of documents,

      a set of topics, and a set of relevance  between documents and topics.

     Accordingly, our task is also separated into three parts. The document set

      is downloaded from various electronic news sites, and totally 132,207

     documents are collected.  To build the topics, we investigate the real

     user information needs by using a questionnaire, and then modify them to

     be the formal topics. As to relevance judgment, we first set up a pool of

      candidate documents for each topic, and then invite three persons to

     judge the relevance. Finally, we combine the judgments and offer a

     relevance measure for each document in the pool. The result of  our

     research shows that the benchmark possesses a complete structure and

     medium scale, and we may further expand and improve it based on existing

     framework in the future.

 

科資中心編號:RB8910-0297

題名:跨語言資訊檢索的網路資源探勘研究

     Mining Internet for Cross Language Information Retrieval

研究者:張俊盛

       Chang, Jyun-Sheng

機構名稱:清華大學資訊科學系(NTHUCOS)

經費來源:行政院國家科學委員會(NSC)

補助金額:新台幣 520 仟元

補助編號:NSC88-2213-E007-054

研究日期:199808-199907

語文:中文

原件提供:行政院國家科學委員會科學技術資料中心 NSC88-2213-E007-054

科資分類號:IA0405;IA0202

關鍵詞:資料探勘;跨語言資訊檢索;多語言翻譯辭典;網際網路;翻譯詞彙集

       Data Mining;Cross Language Information Retrieval;Multilingual

       Translation Lexicon;Internet;Translation Lexicon Construction

摘要:製作跨語言資訊檢索系統,需要多語言的一般辭彙、專有名詞、專業術語等的翻譯詞彙集。透過翻譯詞彙集,系統才能銜接不同語言的資訊需求與內容的目的,翻譯轉換查詢句,檢回的含有相關內容的文件。跨語言的一般辭彙的資源雖然可以由雙語電子辭典取得一大部份。但是語言是活動、靜態的資料總是有時而窮。專有名詞如人名、地名、機構名稱、產品名稱,層出不窮、日新月異,辭典資料很難涵蓋。專業術語的更替,也隨著技術不斷推陳出新,不易掌握。<BR> 目前這些辭彙的建置作業完全都是手工進行,工作煩瑣、重複,十分耗費人力,有必要研發自動化的技術與工具。我們研究在網際網路上進行資源探勘,蒐集跨語言的辭彙資料,作為跨語言檢索研究的基礎。本計畫進行了網頁擷取的自動機、網頁語言的自動決定、網頁互譯性的自動決定、由互譯網頁擷取翻譯詞彙集等項研究。

      The rapid growing Internet has heightened the pressing need of cross

     language information assimilation. One form of such information

     assimilation is cross language information retrieval(CLIR). The  crux of

     building a CLIR system lies in the multilingual translation lexicon

     through which the system can translation queries or retrieved documents to

      bridge the information need and information  content that happened to be

     written in two different languages. Cross language lexical resources for

     common words can be found in large quantity in bilingual machine readable

     dictionaries. However,  keeping up with the ever-changing nature of

     language is always a problem with a static data set. Proper name such as

     person name, place names, organization names and product name that keep

     pouring out day in day out are especially difficult for translators in the

      past and for today's emerging CLIR systems. We carried out a series of

     experiments to meet the new demand for tools that automate the

     acquisition of cross language lexical resources in a speedy fashion. This

     study involved tools as web page collection robot, language identification

      agent, text and translation identification and  automatic extraction of

     translation lexicon from translated web pages.

 

科資中心編號:RB8910-0252

題名:文件分析與網頁文件之自動建立

     Document Analysis and Automatic Construction of Its Relative Web Document

研究者:白敦文

       Pai, Tun-Wen

機構名稱:台灣海洋大學資訊科學系(NTOSCOS)

經費來源:行政院國家科學委員會(NSC)

補助金額:新台幣 391 仟元

補助編號:NSC88-2213-E019-002

研究日期:199808-199907

語文:中文

原件提供:行政院國家科學委員會科學技術資料中心 NSC88-2213-E019-002

科資分類號:IA0407;IA0406

關鍵詞:網頁文件;資訊檢索;文件分析;數位圖書館;文字辨識

       Web Document;Information Retrieval;Document Analysis;Digital Library;

       Character Recognition

摘要:本研究計劃是以智慧型文件分析與文字辨識技術為基礎的網頁文件製作系統,主要是提供紙張文件(書籍或報章雜誌)在成為數位化資料的過程時,圖片與文字部分之自動分離,並且依據其內容之相對位置轉為網頁文件,文字影像部分可以進行文字辨識,圖片影像部分則可以進行部分之重新掃瞄,並提供其相對之註解文字,數位化之網頁文件在超文件標註式語言的描述下,非文字資訊的灰階或彩色圖片,可以藉由超文件標註式語言文件中圖片標籤的"來源路徑"與"替換文字"屬性,使整篇文件在網路上的傳送可以降低對網路頻寬之要求。<BR>      整個計劃的進行中所使用的文件影像資料庫,仍局限在受限制的文件影像(Constrained Documents),而未受限制的文件(Unconstrained  Documents)因為影像區域的黑色連結區塊座標與文字區塊的黑色連結區塊座標交叉重疊,傳統的黑白文件分析技術不易克服其困難度,須藉助彩色文件之輸入或多邊形座標的記錄方式才會有較完整的解決方法,而此技術正研究發展之中。

      This Project provide an intelligent Web Document pagemaker system based

     on document analysis and optical character recognition techniques. The

     main contribution is to reduce the expansive cost price  of labor force

     for digitizing existing paper-based documents in digital library projects,

      and to improve the efficiency and integrity of the network information

     retrieval services. The scanned  documents are classified into two major

     types for Web document translation. One type is text information and the

     other is non-text information which include images and graphs. Text

     information will be obtained directly from OCR, and non-text information

     will be acquired by consequences of document analysis, and its relative

     coordinates and caption information in a document will be recorded in

     order to re-scan the figures by setting scanning parameters with color for

      graylevel attributes.

 

科資中心編號:RB8910-0189

題名:應用模糊積分法於自動資料擷取系統

     Applying Fuzzy Integral in Automatic Documentary Retrieval Systems

研究者:張德民

       Chang, Te-Min

機構名稱:中山大學資訊管理系(NCSNIFN)

經費來源:行政院國家科學委員會(NSC)

補助金額:新台幣 368 仟元

補助編號:NSC88-2416-H110-054

研究日期:199808-199907

語文:中文

原件提供:行政院國家科學委員會科學技術資料中心 NSC88-2416-H110-054

科資分類號:IA0304

關鍵詞:文件擷取系統;模糊積分;模糊量測;概念圖

       Documentary Retrieval System;Fuzzy Integral;Fuzzy Measure;Concept Graph

摘要:資訊爆炸的今日使得人們要掌握正確的資訊必須有賴於資訊擷取系統的協助。但是傳統資訊擷取系統的方式只考慮到使用者查詢個別概念的強度,而忽略概念間的關聯性,本研究利用模糊量測來表達概念間的關聯性,並且以模糊積分法來擷取文件,將可以找到更接近使用者所想要查詢的相關資料。<BR> 同時,在本研究中,我們定義概念圖來幫助我們建構完整的模糊量測,這樣的建構方式的方式可以避免使用者使用系統時因為要輸入太多資訊而感到困擾。

      A good information retrieval (IR) system is essential for us to gather

     relevant information in today's world filled with mass information.

     Traditional IR systems, however, do not consider the  interrelationship of

      users' query concepts, which may result in retrieving less relevant

     information. In this study, we adapt fuzzy measure to capture concepts'

     interrelationship and employ fuzzy  integral as the retrieval function to

     match documents most desired by users.<BR> In addition, we develop a

     concept graph from which we can automatically assign fuzzy measure values.

      In this way,  users only have to provide minimum information regarding

     their query concepts. It makes users easier to operate the system.

 

科資中心編號:RB8910-0170

題名:支援資訊檢索功能的擴充物件資料庫查詢模型及內容索引方法之研究

     Research on an Extended OODBMS Query Model and a Content-Based Indexing

     Method Support for Information Retrieval

研究者:李俊宏

       Lee, Zun-Hon

機構名稱:長榮管理學院資訊管理系(CJUGINM)

經費來源:行政院國家科學委員會(NSC)

補助金額:新台幣 239 仟元

補助編號:NSC88-2218-E309-001

研究日期:199901-199907

語文:中文

原件提供:行政院國家科學委員會科學技術資料中心 NSC88-2218-E309-001

科資分類號:IA0304

關鍵詞:內容索引方法;資訊檢索;物件導向資料庫;查詢模型;文件資料庫

       Content Based Indexing Method;Information Retrieval;Object Oriented

       Database;Query Model;Document Database

摘要:資訊檢索(Information retrieval)功能是文件資料庫的基本需求之一,但現有的資料庫管理系統產品大都無法提供充分的全文檢索(Full-text information retrieval)功能。物件資料庫具有較佳的處理複雜性物件(Complex  objects)能力以及易與其他應用系統結合的長處,故被視為理想的文件資料庫開發及儲存工具;本研究計畫將提出一個使用物件資料庫所開發的文件資料庫系統(Document database system),其中包括一個擴充的物件查詢模型(Query model)及建立內容索引(Content-based indexing )的方法,藉以有效結合物件資料庫管理系統與資訊檢索系統兩者的優點來增進文件資料庫的管理及檢索功能。

      The function of Information Retrieval (IR) is one of the fundamental

     requirements for a document database. However, most available database

     management system (DBMS) products cannot offer sufficient  functions for

     performing full-text information retrieval. Object-oriented database

     management systems (OODBMSs), with the merits that they are excellent at

     handling complex objects and easy to combine with other application

     systems, are considered as an idea developing and storage tool for a

     document database. In this project, an object-oriented database was

     employed to implement a document  database system, including an extended

     object query model and an content-based indexing method, in order to

     effectively combine the merits of OODBMS and IR systems for improving the

     storage and  retrieval functions of the document database.

 

科資中心編號:RB8806-0082

題名:應用延伸式乏晰觀念網路以作乏晰資訊擷取及全球資訊網文件擷取之研究(I)(II)

     Using Extended Fuzzy Concept Networks for Fuzzy Information Retrieval and

     WWW-Based Document Retrieval (I) (II)

研究者:陳錫明

       Chen, Shyi-Min

機構名稱:交通大學資訊科學系(NCTSIFS)

經費來源:行政院國家科學委員會(NSC)

補助金額:新台幣 513 仟元

補助編號:NSC88-2213-E011-084

研究日期:199808-199907

頁冊數:5頁

語文:中文

原件提供:行政院國家科學委員會科學技術資料中心 NSC88-2213-E011-084

科資分類號:IA0407;IA0304;IA0202

關鍵詞:模糊觀念網路;文件擷取;模糊關係;關聯矩陣;關係矩陣;全球資訊網

       Fuzzy Concept Network;Document Retrieval;Fuzzy Relation;Relation Matrix;

       Relevant Matrix;World Wide Web (Www)

摘要:在本計畫中,我們提出一個以延伸式的乏晰觀念網路為基礎的乏晰資訊擷取方法。在一個延伸式的乏晰觀念網路中包含了節點與連結,每個節點表示一份文件或是一個觀念,而節點和節點之間的連結則表示了節點和節點之間相關聯,其並附有一個二元值以表示節點和節點之間的關聯度及以何種關係相關聯。節點和節點之間關聯度的值可以是任意形狀的乏晰數,而不是只能為介於0與1之間的實數值。節點和節點之間的關係可以是乏晰正相關或乏晰負相關等兩種乏晰關係中的任一種。為了增進乏晰推論的時間,我們並利用關聯矩陣及關係矩陣來塑模此一延伸式的乏晰觀念網路。關聯矩陣中的元素表示節點和節點之間的關聯度,關係矩陣中的元素則表示節點和節點之間的關係。另外,我們亦允許使用者的查詢式可以用任意形狀的乏晰數表示,並可以用乏晰正相關或乏晰負相關的方式作查詢,以增加乏晰資訊擷取系統的彈性。另外由於近年來網際網路的盛行,使得資訊擷取系統搜尋符合使用者需求的文件資料所在的範圍也應由單一主機擴展到多主機的網路環境,因此傳統資訊擷取系統已不符合現今網路環境的需求。當使用者在單一主機尋找不到所需的文件資料時,資訊擷取系統必須有能力將搜尋範圍擴充到網路上的其他電腦,直到找到所需的文件資料或其他電腦也沒有所需的文件資料為止。在本計畫中我們亦將原來只適用於單一主機的乏晰資訊擷取方法加以擴展使其成為能適用於網際網路環境的乏晰網路資訊擷取方法。

      In this project, we present a fuzzy information retrieval method based on

      the extended fuzzy concept network architecture. An extended fuzzy

     concept network consists of nodes and links, each node  represents a

     document or a concept, and each link between two nodes associated with a

     tuple represents the relevance between these two nodes, where the tuple

     represents the relevant degree and the  relevant relationship between two

     nodes, respectively. The values of the relevant degree between two nodes

     not only can be real numbers between zero and one, but also can be

     arbitrary shapes of fuzzy  numbers. Moreover, the relevant relationship

     between two nodes not only can be fuzzy positive association, but also can

      be fuzzy negative association. In order to reduce the time of fuzzy

     inference,  we use the relevant matrices and the relation matrices to

     model the extended fuzzy concept network. The elements of a relevant

     matrix represent the relevant degrees between two nodes. The elements of

     a relation matrix represent the relevant relationships between two nodes.

     Furthermore, we also allow users' queries to be represented by arbitrary

     shapes of fuzzy numbers and to use fuzzy positive  association

     relationship and fuzzy negative association relationship for formulating

     their queries for increasing the flexibility of fuzzy information

     retrieval systems. Furthermore, because of the  prevalence of the Internet

      technology, the searching boundary of the information retrieval systems

     for the documents corresponding to user requests should be expanded from

     the local computers to the  network environment of multiple computers.

     When the user can not find the needed documents in the local computer, the

      information retrieval systems should have the ability to search the

     documents in  other computers in the network until the needed documents

     are found or it does not exists. In this project, we also propose a

     computer-network-based fuzzy information retrieval method which is

     suitable for the computer network environment.

 

科資中心編號:RB8806-0070

題名:以概念結構為主的文件自動分類系統

     Automatic Text Categorization Based on Conceptual Knowledge

研究者:柯淑津

       Ker, Sur-Jin

機構名稱:東吳大學資訊科學系(SCUCCIS)

經費來源:行政院國家科學委員會(NSC)

補助金額:新台幣 321 仟元

補助編號:NSC88-2213-E031-003

研究日期:199808-199907

頁冊數:4頁

語文:中文

原件提供:行政院國家科學委員會科學技術資料中心 NSC88-2213-E031-003

科資分類號:IA0411

關鍵詞:概念結構;文件分類;資訊檢索

       Conceptual Structure;Text Categorization;Information Retrieval

摘要:文件分類一直是資訊檢索中相當重要的課題之一。經過分類整理後,主題相關的文件會被分在同一個類別,在後續的檢索處理過程,針對使用者的問題,我們可以先尋找相關的主題類別,然後針對此一分類(或幾個分類),再繼續檢索與問題最匹配的文件。這種作法不需要對所有的文件進行搜尋比對,因此更容易在有限時間內,完成檢索工作,並且達到較精確效果。<BR> 先前以字詞為主的統計式文件分類作法,對於高頻詞通常可以得到高精確度。但是,對於低頻詞彙往往無法得到可靠的機率估算,而造成低涵蓋率的結果。另外,繁複的統計,往往需要耗費大量的運算時間以及儲存空間。透過觀察,我們發現語言的多樣性詞彙變化,往往局限於同一概念詞群中。因此,我們設計一套以概念結構為主的文件自動分類系統,將文件內容由字詞轉化為概念式主題,再依其概念主題進行文件自動分類,這種作法降低了參數的需求量,可以不用仰賴大量訓練資料。為了驗證本計畫所題方法的效能,我們以網際網路上的新聞資料進行測試,將收集自網路上的每日新聞,依其概念分類,分為:國內政治、財貿經濟、台灣社會、大陸港澳、投資理財、影視娛樂、國際新聞、產業科技、體育休閒、氣象交通、生活消費,以及藝文活動等十二種分類。

      Text categorization for unrestricted text is one of the important issues

     in the field of information retrieval. The crux of the problem is to

     discover a model that relates words in a document to its  general subject

     area. It seems to be very difficult to statistically acquire enough word-

     based knowledge to make a robust system capable of automatically

     categorizing unrestricted text. The major  problems with word-based text

     categorization models include data sparseness and the lack of a level of

     abstraction. Word-based text categorization systems are hard to train

     sufficiently well,  furthermore, they are difficult to port to new domains

      and run off the shelf. In this paper, we will show that a concept-based

     model for text categorization requires fewer parameters and has a built

     in element of generality. Broad lexical conceptual knowledge acquired from

      machine readable dictionaries can be used to produce a robust and

     portable text categorization system. A series of  experiments was

     conducted to categorize on-line news obtained from the Internet in order

     to assess the performance of the proposed method. Experimental results

     show that the MRDs function effectively  as a knowledge base for assigning

      subject areas to news articles and for text categorization in general.

 

科資中心編號:RB8803-0147

題名:以類神經網路為基礎的個人化文件擷取系統研究

     The Design of a Personalized Information System Based on Neural Network

研究者:陳培敏

       Chen, Pei-Min

機構名稱:東吳大學資訊科學系(SCUCCIS)

經費來源:行政院國家科學委員會(NSC)

補助金額:新台幣 391 仟元

補助編號:NSC87-2213-E031-001

研究日期:199708-199807

頁冊數:4頁

語文:中文

原件提供:行政院國家科學委員會科學技術資料中心 NSC87-2213-E031-001

科資分類號:IA0407;IA0304

關鍵詞:類神經網路;資訊檢索;個人描述檔;查詢詞;索引關鍵詞

       Neural Network;Information Retrieval;User Profile;Query Term;Index Term

摘要:因每個人對語詞詮釋的不同,故一般資料查詢結果不一定是為查詢者所要的,所以正確又適切地提供資訊給不同需求的查詢者,是現今資訊擷取系統一個非常重要的目標。本計畫設計一個簡易的類神經網路個人化資訊擷取系統,使用者的查詢詞可先經由類神經網路的推衍,找出該查詢最適切的語意關鍵詞,然後進行資訊擷取之動作。但因類神經網路架構(如輸入節點數與輸出節點數等)不易隨時更動,故無法將新增的查詢詞與語意關鍵詞有效的融入元類神經網路中,而造成使用不便,甚至導致擷取資料的誤差。所以此計畫自行設計一個具有語詞相關度回饋調整的機制,以掌握使用者對語詞間彼此語意的關連。

      Providing the right information to the right person in a reasonable time

     is a very important goal for today's information retrieval (IR) system. In

      this study, an adaptive information retrieval system with an embedded

     neural network is proposed to capture the personal notions of query terms

     as well as the degree of relevance among terms. This neural network is

     used as a personal profile to " simulate" the user's notions of term

     semantics. Through the personal profile, the index terms relevant to the

     issued query terms are deduced and they are used in the process of

     information retrieval.  Thus, the system can reduce the discrimination

     between the information needs and the retrieving results. However it is

     not easy to modify the architecture, e.g., to increase the number of nodes

      in the input layer or the output layer, of a neural network after

     constructing. Thus the new query terms and new index terms can not be

     inserted into the neural network such that the IR system becomes not  easy

      to use and even introduces the retrieving errors. In this project, we

     also design a mechanism as a personal profile which possesses the adaptive

      capability to adjust the relevance between terms  after query feedback.

 

科資中心編號:RB8707-0456

題名:以知識庫及聚類分析為基礎的文件擷取系統研究

     The Design of a Document Retrieval System Based on Knowledge Base and

     Cluster Analysis

研究者:陳培敏

       Chen, Pei-Miin

機構名稱:東海大學資訊科學系(THUSCIS)

經費來源:行政院國家科學委員會(NSC)

補助金額:新台幣 367 仟元

補助編號:NSC86-2213-E031-001

研究日期:199608-199707

頁冊數:79頁

語文:中文

原件提供:行政院國家科學委員會科學技術資料中心 NSC86-2213-E031-001

科資分類號:IA0407

關鍵詞:文件擷取;知識庫;聚類分析;模糊集合;概念網路

       Document Retrieval;Knowledge Base;Clustering Analysis;Fuzzy Set;Concept

       Network

摘要:本計畫設計並製作一套以知識庫及聚類分析為基礎的文件擷取系統研究,系統內的文件依其對概念的模糊關係先行分類,再進行擷取之工作。<BR> 在本系統中,專家預先建立的知識庫內容是整個應用領域中各個概念之間的模糊關係以及各個文件與概念間的模糊關係,我們用一個概念網路加以表達,並把該概念網路用二維矩陣儲存,我們稱為概念矩陣。再運用模糊集合理論,將概念矩陣作運算,找出其等價關係矩陣,以充分表達出兩兩概念或兩兩文件之間的關係。<BR> 我們的系統模型分為兩個子系統:一為文件分類、一為文件擷取,這兩個子系統分別在兩個階段時執行。第一階段是離線作業,文件分類子系統以概念矩陣作為輸入資料,進行文件聚類分析,將系統內全部文件分為若干群組;第二階段則為線上查詢作業,文件擷取子系統根據使用者線上鍵入之資料(與查詢文件相關之模糊向量),先搜尋出相關的文件群組,再進一步找出最符合查詢的文件。<BR> 本系統主要是考慮時,讓使用者可以下達模糊概念的查詢指令,並可在大量文件資料庫中快速又正確地找到其欲查詢的文件。另外本系統的效率及準確性也進行初步之分析與評估,以驗證本系統之預期效益。

      A Knowledge-based system model for fussy information retrieval is

     proposed where documents are categorized into clusters based on the degree

      of relevance among them. In this system, a concept network, used as

     knowledge, depicts the relevant relationships among concepts. This concept

      network is represented by a concept matrix, where the implicit relevant

     values between concepts, documents, and both are derived by the transitive

      closure of the concept matrix based on fuzzy logic.<BR> The system model

     is divided into two subsystems: document clustering and document

     retrieving. These two  subsystems are performed off-line, where the

     subsystem of document clustering figures out the document clusters based

     on the concept matrix. The second phase is an on-line job that the

     document  retrieving subsystem retrieves documents from related clusters

     according to user's queries. The proposed system allows users to retrieve

     documents effectively and efficiently through fuzzy queries.  The

     effectiveness and efficiency of the system are also briefly measured and

     analyzed.

 

科資中心編號:RA8507-2076

題名:利用以知識庫為基礎的乏晰資訊尋取技術作文件尋取之研究

     Document Retrieval Using Knowledge-Based Fuzzy Information Retrieval

     Techniques.

研究者:陳錫明

       Chen,Shyi-Ming

機構名稱:交通大學資訊科學研究所(NCTUINSG)

經費來源:行政院國家科學委員會(NSC)

補助金額:新台幣 295 仟元

補助編號:NSC83-0408-E009-041

研究日期:199308-199407

出版日期:199407

頁冊數:42頁

語文:英文

原件提供:行政院國家科學委員會科學技術資料中心 NSC83-0408-E009-041

科資分類號:IA0407

關鍵詞:人工智慧;模糊邏輯;資訊檢索系統;文獻檢索;知識庫

       Artificial Intelligence;Fuzzy Logic;Information Retrieval System;

       Document Retrieval;Knowledge Base

摘要:傳統資訊尋取系統採用布林邏輯模式,但由於必須經常面對乏晰資訊處理的問題,在實用上受到相當程度的限制,無法滿足實際需求。儘管近年來有些乏晰資訊尋取系統相繼被提出,但都著重於不確定資料的表示或資料與查詢項之間相似度的處理,尤其是當面對一大型資料庫而言要逐項進行相似度的比對是相當沒效率的。另外,目前存在的乏晰資訊尋取系統尚不夠聰明,若我們能將人工智慧的技術引進乏晰資訊尋取系統中,則所設計出的乏晰資訊尋取系統將更具智慧。<BR> 在本計畫中,為了達到有效率的管理文件以改善乏晰資訊尋取系統的反應時間,並增加乏晰資訊系統的智慧,我們提出了一個以知識庫為基礎的乏晰資訊尋取模式,其除了能適切的處理資訊之不確定性外,並提供了一個很有效率之乏晰資訊尋取策略以作文件尋取。在本計畫中所提之乏晰資訊尋取技術將實際應用在圖書館之文件查詢上;當給定一組乏晰查詢時,系統將以乏晰邏輯作推理,並以最有效率的方式傳回一組滿足查詢之文件串列。

      Traditional information retrieval systems are founded on Boolean logic.

     Because the fuzzy information processing problem, is often encountered the

      applicability of these systems are restricted and  their performance

     usually can not meet the real requirement. In recent years, some fuzzy

     information retrieval systems have been proposed. However, these systems

     usually focus on the representation of uncertain data or the measure of

     the degree of matching between the data and the query. To perform the

     matching operations between the data and the query sequentially in a large

      database is very  inefficient due to that the existing fuzzy information

     retrieval systems are not smart enough. If we can apply artificial

     intelligence techniques to fuzzy information retrieval systems, then the

     designed system will be more "intelligent".<BR> In this project, we

     propose a knowledge-based approach for fuzzy information retrieval. It can

      not only manipulate the uncertain data properly, but also provide an

     efficient fuzzy information retrieval strategy for document retrieval. The

      knowledge- based fuzzy information retrieval technique proposed in this

     project is used to design a document retrieval system in a library

     environment. Given a query, the system can perform fuzzy reasoning based

     on fuzzy logic to deduce a set of documents which can satisfy the query in

      the most efficient  way.

 

 科資中心編號: RF9104-0127 

題名: 以案件模式為基礎的智慧型網路犯罪蒐尋系統

A Model-Based Crime Information Retrieval System on Internet 

研究者: 陳志誠

Chen, Ji-Chern 

機構名稱: 中央警察大學資訊管理系 

經費來源: 行政院國家科學委員會 

補助金額: 新台幣 280 仟元 

補助編號: NSC89-2416-H015-007 

研究日期: 2000/08 - 2001/07 

出版日期: 2001/07 

原件提供: 行政院國家科學委員會科學技術資料中心, NSC89-2416-H015-007 

科資分類號: SH0307;IB0002 

關鍵詞: 電腦犯罪 ;智慧型代理人 ;資訊檢索 ;資料過濾 ;資料分類 ;資料庫建置

Computer Crime ;Intelligent Agent ;Information Retrieval ;Data Filtering ;Data Classification ;Database Construction 

中文摘要: 有鑑於資訊科技相關之犯罪問題日益嚴重,雖然網際網路提供各種大量的資料,但缺乏一個專門蒐集與提供電腦犯罪相關案例報導之服務,造成電腦犯罪偵查或防治在實務上的不便與不足,因此,本研究實作一個以代理人(Agent )為基礎的電腦犯罪資訊提供系統,運用智慧型代理人(IntelligentAgent)機制來完成電腦犯罪案例之蒐集,以節省人力、時間,並提供使用者一個與電腦犯罪相關的整合性資訊服務,企盼能讓系統的維護負荷降至最低;目前,本系統採免費會員制,已蒐集了上千筆犯罪資料,包括國內外及中國大陸等地區,從1997年到今日有關電腦犯罪的案例,共分十二類,另外,也提供相關報導及案例的後續發展、確定判決,希望能提供學者、執法人員或有興趣之人員參考使用。本研究探討智慧型代理人之自動化搜集系統,在整個資料取得的過程中,除了最後須由人工決策最終資料取捨外,其餘包括資料搜尋、過濾..等程序皆藉由智慧型代理人來自動完成,本系統以IR、Filter and Category、Interface、Mail Delivery 代理人來描述各種機制,藉由這些機制與網際網路環境,提供使用者一個整合性的電腦犯罪資料庫服務。 

英文摘要: The problem caused by computer crime becomes more and more serious. Though the Internet is abundant of data, but it is difficult to find information concerning computer crime. Therefore, this research aims at setting up a web station, the InfoCrime system, for providing information of computer crime with the hope of assisting researchers to study and investigate computer crime. The data contained in the InfoCrime system were originally collected from newspapers and journals and input into the system manually. This is often incomplete and labor-intensive. In order to automate data collection, we implement the system based on intelligent agents that perform various functions automatically. Main features of the agents include data searching, data collection, data filtering, and data Analysis . This saves us time and efforts significantly. In addition, the system provides users with an integrated push-service of related information about computer crime cases. The cost of system maintenance is reduced by the agents that automate processes including data storage and data organization. In summary, the agents of the InfoCrime system, that are the IR-, Filter-, Category-, Interface- and Mail-Delivery-agent, work together to provide the functions mentioned above. To the present day, there are over one thousand computer crime cases contained in the system. With this large volume of information, the InfoCrime system is able to accomplish the following achievements: (1) It provides a research environment for the information community to study the function and usage of a non-profit Internet content provider (ICP);(2) It helps law enforcement agencies to investigate computer crime;(3) It provides information for the socio-psychological community to study the behavior of computer criminals; and, not the last, (4) It helps families to educate their children the correct use of the Internet. 

 

科資中心編號: RB9104-0071 

題名: 建立一個在網際網路上執行的分散式計算平台

Constructing a Distributed Computing Platform Based on Internet 

研究者: 王豐堅

Wang, Feng-Jian 

機構名稱: 交通大學資訊工程研究所 

經費來源: 行政院國家科學委員會 

補助金額: 新台幣 750 仟元 

補助編號: NSC89-2213-E009-138 

研究日期: 2000/08 - 2001/07 

出版日期: 2001/12 

頁冊數: 10頁 

原件提供: 行政院國家科學委員會科學技術資料中心, NSC89-2213-E009-138 

科資分類號: IA0407;IA0202 

關鍵詞: 網際網路 ;分散式計算平台 ;智慧型代理人 ;行動代理人 ;通訊服務系統 ;名稱服務系統 ;通訊服務 ;伺服系統

Internet ;Distributed Computing Platform ;Intelligent Agent ;Mobile Agent ;Communication Serivce System ;Naming Serivce System ;Communication Service ;Servo System 

中文摘要: 代理程式(agent programming)模式在過去幾年來日益發達,在網路上移動的代理程式一般稱為移動型代理執行緒(mobile agent),它與單點上的智慧型代理執行緒(intelligentagents)及程式導向的代理執行緒(programorientedagents)有可能成為未來程式撰寫與執行的主流。主要原因是它攜帶部分程式,具主動性質,可在網際網路上自行達成單一或多項目標。本期的工作主要是設計製作一個允許移動行代理執行緒交互溝通的平台。 

英文摘要: Mobile agent paradigm is getting popular in the past few years. Agent programs communicate with each other or local user to accomplish its goal. Mobile agent(program)s roam in the network, and, there needs a mechanism to maintain the location of mobile agents in inter-network. It is difficult to find a target agent. Besides, Intelligent agent(program)s can learn continuously to modify its intension(behavior) based on the belief and desire. Our work describes a new design and the implementation of communication system for our mobile agent system in order for agents to collaborate with each other more effectively. Our communication system includes both a new name service system and a new message delivery based on standard naming/communication mechanisms. The name service system works with location tracking, while the message delivery system handles message deliveries. 

 

科資中心編號: RB9102-0989 

題名: 以代理人為基礎的數位圖書館的資訊搜集與服務

Agent-Based Information Gathering and Service for Digital Library 

研究者: 蘇豐文 ; 蔣嘉寧 ; 劉吉軒

Soo, Von-Wun ; Liu, Jyi-Shane 

機構名稱: 清華大學資訊工程系 

經費來源: 行政院國家科學委員會 

補助金額: 新台幣 1122 仟元 

補助編號: NSC89-2750-P007-002-4 

研究日期: 2000/08 - 2001/07 

出版日期: 2001/10 

頁冊數: 5頁 

原件提供: 行政院國家科學委員會科學技術資料中心, NSC89-2750-P007-002-4 

科資分類號: IA0407;PA0203 

關鍵詞: 智慧型代理人 ;本體論 ;資源描述架構 ;資訊檢索 ;資訊服務 ;索引典 ;加註系統 ;多媒體

Intelligent Agent ;Ontology ;Resource Description Framework (Rdf) ;Information Retrieval ;Imformation Service ;Thesaurus ;Annotation System ;Multimedia 

中文摘要: 本計劃擬設計發展一個以「智慧型代理人」為基礎的數位圖書館的資訊服務系統。此系統以電子化資料為主體,擴及其他資訊源的相關資訊自動搜集與彙整,再加上接受委託完成後續資訊蒐集工作或監看特定資訊源,主動通知回報最新工作成果與消息。我們將利用現有的資訊查詢系統,透過「智慧型代理人」與使用者的互動提供搜尋指引,過濾無用的資訊,達到更精準的查詢。我們將發展不同功能的「智慧型代理人」組成的數位圖書館資訊服務系統。此系統包括下列代理人之設計:「服務代理人」、「領域知識代理人」、「詞彙代理人」、「資訊搜集代理人」等,彼此分工合作,建立不同資訊系統間對話、互動、與彙整的能力,提供跨系統的全面性資訊服務。第一年以建構歷史圖片之加註系統,整合詞彙、領域知識與加註資料。並以資源描述架構語言(RDF)為主要知識表示法。 

英文摘要: In this project, we design agent-based information service systems for digital library. Based on current legacy information retrieval system, intelligent agents will interact with users and provide them information gathering instructions, filtering unnecessary information, to achieve more precise information retrieval. We develop the following information agents: service agent, ontology agent, thesaurus agent, and information gathering agents who could collaborate together. We shall use intelligent agents to establish the interactions and integration among different information systems in order to achieve full-fledge interoperable information service among heterogeneous information resources. In the first year, we establish an annotation system to integrate the thesaurus, ontology and metadata, and use resource description framework (RDF) as the major language to represent ontology. 

 

 

科資中心編號: RF9102-0453 

題名: 理解基礎幾何學的自然語言教育代理人

Natural Language Educational Agent That Understands Elementary Geometry 

研究者: 黃永廣

Wong, Wing-Kwong 

機構名稱: 雲林科技大學電子工程技術系 

經費來源: 行政院國家科學委員會 

補助金額: 新台幣 677 仟元 

補助編號: NSC89-2520-S224-003 

研究日期: 2000/08 - 2001/07 

出版日期: 2001/07 

原件提供: 行政院國家科學委員會科學技術資料中心, NSC89-2520-S224-003 

科資分類號: AA0301;SC0103;IA0407 

關鍵詞: 自然語言理解 ;網站教育代理人 ;幾何教育 ;問題解決 ;智慧型代理人 ;人工智慧

Natural Language Understanding ;Website Educational Agent ;Geometric Education ;Problem Solving ;Intelligent Agent ;Artificial Intelligence 

中文摘要: 當老師或學生編輯網頁教材的時候,他們通常遇到畫圖形的問題。目前的解決辦法常常是利用繪圖工具先畫圖,然後存成bmp 或gif 等圖檔格式,再利用html語法把圖檔在入網頁。可惜這個辦法既耗時、又很難用精準的方法去繪圖。這個研究計畫提出一個網站教育代理人的設計,讓網頁教材作者用自然語言去描述幾何圖形。這個代理人由三個主要元件組成,第一個是Geometry Agent ,負責把自然語言理解成一種中文物件導向描述語言(COOL)。第二個元件是COOL 伺服器,負責將COOL 描述的圖形的屬性計算出來,然後產生可畫出這圖形的CLogo程式。第三個元件是CLogo 視覺化環境,負責執行CLogo 程式去把圖形畫出來。 

英文摘要: When a teacher or student wants to author Web pages of instruction materials, they usually face the problem of drawing figures of geometric objects. The current common approach is to draw the figures with tools to produce images in formats of bmp, gif, etc., which are then included in the Web pages using the Hypertext Markup Language(HTML). This can be time-consuming and difficult to produce figures with great precision. This project proposes to provide an intelligent agent that understands Web page authors who describe their figures of geometric objects in Chinese. The agent consists of three components. The first one is the Geometry Agent that understands the texts as some object-oriented programs written in the Chinese Object-Oriented Language. The second is the COOL server that interprets the COOL programs, solves equations about the geometric objects, and produces CLogo programs that draw the objects. The third component is the CLogo visualization environment that executes the CLogo programs and draws the objects. 

 

科資中心編號: RN9102-0053 

題名: 以網際網路為基礎所發展之分散式智慧型物件導向監控系統

Application of Distributed Web-Based Intelligent and Object-Oriented Monitoring and Control System 

研究者: 郭重顯 

機構名稱: 長庚大學機械工程系 

經費來源: 行政院國家科學委員會 

補助編號: NSC89-2212-E182-013 

研究日期: 2000/08 - 2001/07 

出版日期: 2001/10 

頁冊數: 5頁 

原件提供: 行政院國家科學委員會科學技術資料中心, NSC89-2212-E182-013 

科資分類號: IB0203;IF0302;IE0603 

關鍵詞: 分散式監控系統 ;彈性製造系統 ;智慧型代理人模組 ;分散式彩色時序裴式圖 ;網際網路 ;性能評估

Distributed Monitoring And Control System ;Flexible Manufacturing System (Fms) ;Intelligent Agent Module ;Distributed Colored Timed Petri Net (Dctpn) ;Internet ;Performance Evaluation 

中文摘要: 本計畫是以網際網路為基礎來發展智慧型、物件導向分散式監控系統。此一研究將以分散式軟體元件、智慧型代理模組、分散式單元控制器、網路科技和網頁視窗監控程式等技術和理論來完成本計畫所提出之架構。在本計畫中,彈性製造系統是以階層式的架構來建構智慧型物件導向監控系統。然而,此一架構是以分散式智慧型代理模組來達成彈性製造系統各階層控制器的容錯和負載均衡特性。分散式智慧型代理模組具有分散式地執行自動處理更新資料、自我決策和通訊的能力。在智慧型決策部分採用了分散式彩色時序裴氏圖的理論,其以事件驅動的方式來處理系統中同時發生且不同步的事件,並可使用其資源競爭和衝突解決特性來實現控制器邏輯。最後,本系統將以微軟的Visual C++和SQL Server開發視窗程式來實現本計畫的架構。 

英文摘要: This project proposes an intelligent and object-oriented monitoring and control system, and it is designed for a flexible manufacturing system (FMS). Several technologies, including, distributed software components, intelligent agents, distributed cell controllers, network technology and web-based monitoring programs, complete this research. The FMS intelligent monitoring and control system is constructed based upon the hierarchical architecture. Especially, such a system is developed in terms of distributed intelligent agents. Distributed intelligent agents can automatically update data , and it has the capacities of decision-making and communication. In this research, a colored timed Petri net (DCTPN) is used to implement the intelligent decision making of cell controllers, and it is powerful to handle the asynchronous and concurrent events of the event-driven system. In addition, the resource competition and conflict resolution properties of the DCTPN can be used to realize controller logic. Finally, we use the Microsoft Visual C++ and SQL Server to implement the proposed architecture. 

 

 

科資中心編號: RF9101-0464 

題名: 以智慧型代理程式為基礎的網路議價機制之研究

An Intelligent Agent-based Bargaining System on the Internet 

研究者: 李富民

Lee, Fu-Ming 

機構名稱: 朝陽科技大學資訊管理系 

經費來源: 行政院國家科學委員會 

補助金額: 新台幣 267 仟元 

補助編號: NSC89-2416-H324-024 

研究日期: 2000/08 - 2001/07 

出版日期: 2001/09 

原件提供: 行政院國家科學委員會科學技術資料中心, NSC89-2416-H324-024 

科資分類號: IB0002 

關鍵詞: 智慧型代理人 ;網路議價 ;電子交易 ;電子商務 ;網路交易 ;模糊理論

Intelligent Agent ;Internet Bargaining ;Electronic Transaction ;Electronic Commerce ;Internet Transaction ;Fuzzy Theory 

中文摘要: 由於全球資訊網以爆炸式的速度成長,網際網路使用者急劇擴增,同時也改變了傳統的商務型態,因此,企業紛紛投入電子商務的開發行列,希望透過電子商務,延伸傳統商務的功能,使企業更具競爭力。近年來,電子商務環境逐漸產生複雜性與競爭白熱化,其中以提供價格隨著供需雙方自由決定的『變動價格( Dynamic price )』服務機制最被重視,而議價機制就是『變動價格』的經營方式。但是由於目前的議價機制仍不夠完善,造成買賣雙方仍舊必須在線上進行產品的價格協商,徒增雙方的困擾,也增加了議價的成本。為此,對於網路議價所面臨的問題,本研究考慮以代理程式協助買賣雙方的方式解決。期望透過代理程式的協助,增進網路議價的利用與便利性。 

英文摘要: Electronic commerce (EC) activities are growing rapidly on the Internet. One area that receives much attention is that of bargaining. This activity can become a major EC factor due to the creation of special online brokering services for buyers and sellers.In this project, we propose an intelligent agent for each seller to negotiate the price of good in electronic market, and propose an intelligent agent for each buyer to buy goods or services provided in the electronic market.The purposes of the project are to design a generalized multi-agent market framework that can examine the effect of bargaining in electronic market. This project has great potential to activate the electronic market, to reduce transaction costs for buyers and sellers, to promote the use of the Internet, and to create EC developers. 

 

科資中心編號: RF9101-0461 

題名: 應用智慧型購物幫手於顧客化產品的網路行銷及最佳的訂貨策略的訂定---以個人電腦為例

The Use of Intelligent Shopping Assistant for the Customized Products on the Internet Marketing and the Related Optimal Ordering Strategies 

研究者: 陳大正 ; 游鵬勝

Chen, Ta-Cheng ; You, Peng-Sheng 

機構名稱: 長榮管理學院資訊管理系 

經費來源: 行政院國家科學委員會 

補助金額: 新台幣 196 仟元 

補助編號: NSC89-2416-H309-015 

研究日期: 2000/08 - 2001/07 

出版日期: 2001/07 

原件提供: 行政院國家科學委員會科學技術資料中心, NSC89-2416-H309-015 

科資分類號: IB1101 

關鍵詞: 電子商務 ;網路行銷 ;網路購物 ;智慧型代理人 ;顧客化行銷 ;資訊業 ;個人電腦 ;訂貨策略

Electronic Commerce ;Internet Marketing ;Internet Shopping ;Intelligent Agent ;Customized Marketing ;Information Industry ;Personal Computer ;Ordering Strategy 

中文摘要: 本研究計劃主要探討如何在企業與消費者間的電子商務中的網路行銷上利用購物幫手(shoppingassistants)或稱為代理者(agents)依照個別客戶或使用者對於所欲購買產品不同需求的條件的限制下,而能提供其最符合顧客需求之產品,更進一步考量到企業與企業間的電子商務中的訂貨決策,即找出最佳的訂貨策略也同時被考慮而欲使期望利潤最佳。以往的網路行銷只具單向的溝通,鮮少有互動或雙向的溝通,然而如何找出符合顧客需求的產品或服務一直是「成功的將供應者的產品或服務轉移為顧客的需要」關鍵要素之一,而欲達此目的是利用介於供應商和顧客之間仲介者(intermediary)的功能,其能比顧客本身更有效率的執行找尋符合顧客需求的功能,以擁有更多的競爭優勢。而本研究所需克服的困難包括如何建立一個智慧型的代理者能找出一個產品其能對應到顧客之特定需求。另外是在整合所有顧客之訂單的狀況下,來找出最佳貨品的訂購策略。因此在本計劃中,我們建構一智慧型購物幫手的方法其能及時提供顧客化的產品資訊且能有效找出符合顧客需求的產品,並提出一最佳的訂貨策略,以達到「企業與企業」及「企業與顧客」於顧客化產品的網路行銷及訂貨的電子商務。 

 

英文摘要: This project is to investigate how to apply shopping assistants(agents) for the purpose of marketing on the Internet. The uses of agents in the E-commerce is both for the business to business and business to customer. What appears to the user as one intelligent shopping assistant is actually multiple assistants that simultaneously(in parallel) collect (pull) information from relevant Web sites located worldwide. The purpose of the research is wished to reach two objectives. The first one is to find the customized product for the customer through the Internet based on customers’ needs. The second one is to provide the best ordering policy for the intermediary which will be included in the electronic ordering system. In this project, an intelligent shopping assistant (intelligent agent) is used to find the customized product based on the customer’s needs and the budget through the internet and also, it is able to make the best ordering policy for the intermediary (E-Shop) to its suppliers. In this research, we try to propose a methodology to successfully sell the customized product on the internet for the purpose of E-commerce. 

 

科資中心編號: RB9101-0190 

題名: 以概念圖形法來達成智慧型代理人之知識分享

Conceptual Graphs as a Basis for Knowledge Sharing among Agents 

研究者: 潘健一 ; 賴聯福

Lai, Lein F. 

機構名稱: 建國技術學院資訊管理科 

經費來源: 行政院國家科學委員會 

補助金額: 新台幣 288 仟元 

補助編號: NSC89-2218-E270-002 

研究日期: 2000/08 - 2001/07 

出版日期: 2001/09 

頁冊數: 5頁 

原件提供: 行政院國家科學委員會科學技術資料中心, NSC89-2218-E270-002 

科資分類號: IA0407 

關鍵詞: 概念圖形法 ;知識共享 ;智慧型代理人 ;知識交換格式 ;代理人溝通語言 ;本體論

Conceptual Graph ;Knowledge Sharing ;Intelligent Agent ;Knowledge Interchange Format (Kif) ;Agent Communication Language (Acl) ;Ontology 

中文摘要: 近年來,隨著網路的蓬勃發展,智慧型代理人的研究逐漸地受到相當大的重視,而智慧型代理人的發展成果也被廣泛地應用在各種不同的領域上。智慧型代理人之間經常需要透過互相合作或交互溝通以達成它們的任務與工作,由於每個智慧型代理人根據它們本身不同的特性,可能會使用不同的知識表示語言、不同的推理機制、甚至對某些名詞的定義與解釋都不同,因此,智慧型代理人之間的溝通語言與知識分享成為一個很重要的研究課題。本計畫提出一個使用概念圖來達成智慧型代理人之間知識分享的方法。首先,發展一個知識共通語的階層式架構,來減輕撰寫各種不同版本概念圖與KIF之間轉換程式的成本,以及減少轉換過程中資訊流失的情況。其次,為了達成智慧型代理人之間的知識分享與互相溝通,將針對知識共通語、代理人溝通語言、和知識本體三個部分,整合成一致性的溝通環境,使得所有以概念圖為知識表示語言的智慧型代理人都可以透過此溝通環境來分享知識。最後,為了降低建構知識庫所需的昂貴成本,將利用概念圖建立一個可重複使用的共同知識本體,使得各個知識庫系統,可以透過重複使用他人的知識本體來建立自己本身的知識庫。 

英文摘要: Recently, with the growth of Internet, intelligent agents have received a lot of attention and being used in an increasingly wide variety of applications. Collaboration and communication are necessary for intelligent agents to achieve their tasks. However, different agents may require different knowledge representations, different reasoning mechanisms, and different conceptualization to support them. Knowledge sharing among agents is thus becoming more important. In this project, we utilize conceptual graphs as a basis for knowledge sharing among agents. First, an interlingua hierarchy is developed to reduce the cost of translation between KIF and CGs, and to avoid the loss of partial information. Second, we propose an approach to communicating and sharing knowledge among agents based on the integration of interlingua, agent communication languages, and ontology. Finally, a reusable common ontology in CGIF is established for the purpose of reusability. Building a knowledge base from reusing existing ontology could reduce the cost of constructing from scratch. 

 

 

科資中心編號: RF8910-0537 

題名: 網路商店策略性彈性定價系統之研究

Study of Strategic Flexible Pricing System of a Web Store 

研究者: 賴香菊

Lai, Hsiang-Chu 

機構名稱: 中山大學資訊管理系(NCSNIFN) 

經費來源: 行政院國家科學委員會(NSC) 

補助金額: 新台幣 448 仟元 

補助編號: NSC88-2416-H110-045 

研究日期: 1998/08 - 1999/07 

原件提供: 行政院國家科學委員會科學技術資料中心 NSC88-2416-H110-045 

科資分類號: SM06;IB0002;IB1100 

關鍵詞: 網路商店 ;定價策略 ;智慧代理人 ;電子商務 ;網際網路 ;行銷策略 ;網路行銷 ;彈性訂價

Internet Store ;Pricing Strategy ;Intelligent Agent ;Electronic Commerce ;Internet ;Marketing Strategy ;Internet Marketing ;Flexible Pricing 

中文摘要: 近年來隨著網路的商業應用已經有如野火般的蔓延開來;但一般的網路商店在產品價格方面無法針對不同情形給予最適當的報價。本計劃以傳統的定價策略配合網路的特性,並結合智慧代理人的應用,提出一個線上彈性定價的系統架構。此外本計劃採用四種不同的彈性定價方法,以實驗的方式來對其行銷效果加以衡量。 

英文摘要: Advances of internet and world wide web technology have made electronic commerce possible and increasingly important. Many organizations have actively participated in such an electronic marketspace to gain strategic advantages or create business opportunities. Currently the uniform pricing strategy was commonly observed in a webstore. This research aims at extending traditional pricing strategies by considering the characteristics of internet. A system architecture for an online flexible pricing system based on the intelligent agent technology will be proposed. The preliminary experimental result shows that different flexible pricing strategies have a significant effect on the purchasing quantity and the number of webpages browsed. 

 

科資中心編號: RC8604-1256 

題名: 運用人工智慧支援決策模式建構之研究

Applying Artificial Intelligence to Support the Construction of Decision Models. 

研究者: 梁定澎

Liang,Ting-Peng 

機構名稱: 中山大學資訊管理系(NCSNIFN) 

經費來源: 行政院國家科學委員會(NSC) 

補助金額: 新台幣 534 仟元 

補助編號: NSC85-2416-H110-017 

研究日期: 1995/08 - 1996/07 

原件提供: 行政院國家科學委員會科學技術資料中心 NSC85-2416-H110-017 

科資分類號: SM0404;IB0205 

關鍵詞: 決策支援系統 ;模式管理 ;模式建構 ;人工智慧 ;智慧代理人

Decision Support System ;Model Management ;Model Construction ;Artificial Intelligence ;Intelligent Agent 

中文摘要: 企業決策中有許多涉及複雜的因果關係,需要良好的決策支援系統。決策模式的運用與管理則是決策支援系統有效運用的關鍵性因素之一,因此大部分決策支援系統的開發與應用都需要有決策模式來協助。在決策支援系統中協助模式運用的子系統通常稱之為模式管理系統。本研究之主要目的在探討運用人工智慧建構模式管理系統之相關研究,並尋找未來之研究方向。我們首先利用一個架構,來分析目前運用人工智慧進行模式管理的研究現況。再據以找出目前研究有所不足而具有潛力的未來研究方向。最後,針對智慧代理人在未來的發展潛力,我們特別加以深入討論。 

英文摘要: Models are critical to complicated business decisions. Model management systems (MMS) that facilitate the use of decision models are a major research focus in decision support systems (DSS). The primary purpose of the study is to investigate how artificial intelligence (AI) techniques have been used to design MMS. We develop a framework to analyze AI applications in model management . Existing research and promising future directions are categorized. An approach that uses intelligent agents to develop a distributed model management systems (DMMS) is also proposed.  

 

 

科資中心編號: RB9102-0472 

題名: 具安全性考量的智慧型會議秘書之設計與實作

Design and Implementation of an Intelligent Conference Agent with Security Considerations 

研究者: 楊豐兆 ; 曹偉駿

Yang, Feng-Chao ; Tsaur, Woei-Jiunn 

機構名稱: 大葉大學資訊管理系 

經費來源: 行政院國家科學委員會 

補助金額: 新台幣 300 仟元 

補助編號: NSC89-2213-E212-031 

研究日期: 2000/08 - 2001/07 

出版日期: 2001/07 

原件提供: 行政院國家科學委員會科學技術資料中心, NSC89-2213-E212-031 

科資分類號: IA0202;IA0407 

關鍵詞: 會議代理人 ;知識管理 ;智慧型代理人 ;安全性考量 ;網際網路 ;知識共享

Conference Agent ;Knowledge Management ;Intelligent Agent ;Security Consideration ;Internet ;Knowledge Sharing 

中文摘要: 本計畫是在網際網路(Internet)環境下,採視訊會議(e-conference)技術與知識管理(Knowledge management, KM)理論,設計一具知識分享式(Knowledge sharing)的智慧型會議秘書,並以辯論比賽為例說明本系統的使用。本系統將現有的視訊會議系統、知識管理系統與擬人化智慧型代理人相結合,可以協助辯論比賽會議流程與程序、資料的搜集、議程的管理、辯論佐證資料的提供、辯論後的檢討及追蹤工作,以提昇辯論整體的效益。 

英文摘要: In this project an intelligent conference agent with knowledge sharing will be designed. It is a convenient and effective tool for e-conference. Based on the intelligent agent and cooperative distributed problem-solving techniques, the main aim of this software system is to integrate the knowledge management system into e-conference to assist users to enhance the performance and efficiency of e-conference. 

 

科資中心編號: RB9101-0987 

題名: 以智慧型代理人技術為基礎之生產規劃與控制決策支援系統

Design of An Intelligent Agent-Based Decision Support System for Production Planning and Control 

研究者: 王孔政

Wang, Kung-Jeng 

機構名稱: 中原大學工業工程系 

經費來源: 行政院國家科學委員會 

補助金額: 新台幣 270 仟元 

補助編號: NSC89-2213-E033-058 

研究日期: 2000/08 - 2001/07 

出版日期: 2001/09 

原件提供: 行政院國家科學委員會科學技術資料中心, NSC89-2213-E033-058 

科資分類號: IB0401;IB0002 

關鍵詞: 智慧型代理人 ;生產規劃 ;決策支援系統 ;生產排程

Intelligent Agent ;Production Planning ;Decision Support System ;Production Scheduling 

中文摘要: 本研究的目標在發展一個「以智慧型代理人技術為基礎之生產規劃與控制決策支援系統」。輔助企業既有的生產規劃與控制系統,可在具風險性環境下進行更有效的生產規劃與控制,對既有系統的衝擊與再投資也減至最小。所發展的系統,可用協調方式,使分佈多處、生產資訊互通不易、或有資訊時差、甚或其命令系統亦不盡相同的生產系統,以協同方式,完成繁雜訂單之生產規劃與控制。本研究所發展的生產規劃與控制決策支援系統,具有智慧型、自主性、自我調整、合作性、可組態、及可延伸等特色。代理人可調整本身的決策行為,以求完成目標。對於劇變的市場變化與未可預知的動態系統狀態,預期比傳統程序式預設的生產規劃與控制模式優越。本計劃之主要研究項目如下:1.以智慧型代理人技術為基礎的決策支援資訊系統架構2.智慧型代理人群組產能分派3.以協商決策函數模型進行智慧型代理人間生產排程本研究所發展的生產規劃與控制決策支援系統,解決半導體測試廠多資源並用之產能分派與排程之研究議題。 

英文摘要: The goal of this research project is to develop an intelligent agent-based decision support system for production planning and control. The system enables efficient planning and control through its decision support function which is developed for risky environment. The system allows intelligent agents in the system to coordinate among each others to complete sophisticated capacity planning and production control. The focuses of the research are as follows. l Information infrastructure of decision support system using intelligent agent technology; l Task allocation among intelligent agents using distributed searching; and l Production scheduling using negotiation decision function model. Two significant issues in semi-conductor industry are proposed to be solved: l Concurrent control of job scheduling, transportation, and dispatching in wafer fabrication factories; and l Multi-resource planning incurring in wafer testing factories. 

 

科資中心編號: RB9101-0641 

題名: 基於智慧型代理技術的個人化教學網頁設計

Personalized Teaching Homepages Design Based on Intelligent Agent Technologies 

研究者: 王勝德

Wang, Sheng-Der 

機構名稱: 台灣科技大學電機工程系 

經費來源: 行政院國家科學委員會 

補助金額: 新台幣 709 仟元 

補助編號: NSC89-2218-E002-030 

研究日期: 2000/08 - 2001/07 

出版日期: 2001/07 

原件提供: 行政院國家科學委員會科學技術資料中心, NSC89-2218-E002-030 

科資分類號: IA0407;IA0408 

關鍵詞: 智慧型代理人 ;網路教學 ;個人化教學網頁 ;可擴展標示語言 ;爪哇 ;延展性 ;電腦輔助教學 ;網頁設計

Intelligent Agent ;Network Teaching ;Personalized Teaching Homepage ;Extensible Markup Language (Xml) ;Java ;Scalability ;Computer Assisted Instruction (Cai) ;Web Page Design 

中文摘要: 網路的進步讓我們的生活和網路更緊密的結合,本計劃就是要探討這樣的線上學習環境,目前已經有很多線上的學習系統,不過都和傳統的教學模式雷同,因此仍然很多的改進空間。本計劃設計並實作了一個分散式環境下多智慧型代理程式的個人導向線上學習系統,針對傳統的教學模式的缺點,系統強調以使用者的觀點,讓系統來符合使用者的需求,讓系統了解每個使用者的差異性,在使用者使用系統時完全以使用者的需求及進度來設計系統,真正達到因材施教的目標。我們提出了一個多智慧型代理程式的平台架構。在這個架構上,我們使用跨平台的 Java 程式語言,與具延展性,可重複利用的XML 文件格式,提供智慧型代理程式彈性的進出系統機制、以智識為訊息交換內容,及提供智慧型代理程式對等(Peer to peer)的通訊功能。開發者只需要專注於智慧型代理程式的特殊功能之上。其他例如與使用者溝通,或者是尋求協助等等,都透過系統所提供的相關服務完成。雖然這個系統是以教學系統來發展,不過這樣的架構不只限於教學系統的應用。目前很多線上的商業軟體都可以採用這樣的架構。在抽換不同的模組,適當的修改,可以讓系統應用在不同的領域。 

英文摘要: The research intends to take advantage of intelligent agent technologies to design and implement personalized teaching homepages, and a ubiquitous, personalized, networked educational system as well. Our design is based on multi-agent systems that are then making use of the XML and Java technology. The system provides an intelligent agent platform such that portable agents can easily exchange their ontology and communicate peer to peer. This model promotes the scalability of our system. Finally, the design and implementation issues are also addressed in this report. 

 

科資中心編號: RB8910-0486 

題名: 設計整合式智慧型品質機能展開系統

The Design of an Integrative Intelligent QFD System 

研究者: 高信培

Kao, Hsing-Pei 

機構名稱: 中央大學企業管理研究所(NCUGBUAG) 

經費來源: 行政院國家科學委員會(NSC) 

補助金額: 新台幣 352 仟元 

補助編號: NSC88-2213-E008-026 

研究日期: 1998/08 - 1999/07 

原件提供: 行政院國家科學委員會科學技術資料中心 NSC88-2213-E008-026 

科資分類號: IB0500;IA0302 

關鍵詞: 品質機能展開 ;智慧型代理人 ;物件導向模型 ;統一模式語言 ;績效評估 ;衝突解決

Quality Function Deployment (Qfd) ;Intelligent Agent ;Object Oriented Model ;Unified Modeling Language (Uml) ;Performance Evaluation ;Conflict Resolution 

中文摘要: 產品開發過程中,應用品質機能展開法(Quality function deployment, QFD)能將市場的需求有效地轉化成產品的功能特性、設計需求與規格、製造程序與控制參數、生產規劃與排程、成本估算等資訊,使得產品研發設計的整合工作與品質管理更為徹底有效。本研究所發展的整合式智慧型品質機能展開法的系統雛型主要係以QFD作為分析架構設計多代理者網路黑板(Multiagent-based blackboard),以提供分散式計算環境(企業內網路)的合作機制。本研究的具體成果包括:(1)建立產品設計、組裝、製造的評估機制發展智慧型展開的演算法,(2)發展智慧型偵測及解決設計衝突的演算法,(3)發展智慧型決策支援的演算法,(4)以物件導向的 UML 分析與設計I2QFD系統以及(5)以實例驗證所發展的方法。 

英文摘要: With a structured process and a set of interrelated engineering and management matrices, QFD ensures the design quality while the product is still in the design stage. Following the deployment process , various functions (i.e., marketing, product design/engineering, prototype evaluation, production planning, manufacturing, and sales) collaborate and communicate to achieve a more balanced overall performance in terms of customer satisfaction, manufacturability/assemblability, logistics, cost, etc. Considering the implementation of QFD on the Intranet, this study developed an integrative intelligent QFD system (I2QFD) which is specifically designed into a blackboard platform, to support asynchronous integration. Through I2QFD, various sources of information can be recorded, synthesized, evaluated, and transmitted. To provide necessary functions, such as functional deployment, conflict detection and resolution, and decision support, several computational/intelligent agents (C/IAs) are developed for the I2QFD. We also apply well-known axioms of minimum information and design independence as well as some suitable DFMA evaluation methods into the system to support design validation. For modeling the I2QFD framework, an object-oriented analysis and design language: UML is utilized. 

 

科資中心編號: RB8910-0416 

題名: Web-Based虛擬商店之研究---網路商店之運作(II)

Study on Web-Based Electronic Store---Operations of Electronic Store(II) 

研究者: 陳年興

Chen, Nian-Shing 

機構名稱: 中山大學資訊管理系(NCSNIFN) 

經費來源: 行政院國家科學委員會(NSC) 

補助金額: 新台幣 340 仟元 

補助編號: NSC88-2416-H110-055 

研究日期: 1998/08 - 1999/07 

原件提供: 行政院國家科學委員會科學技術資料中心 NSC88-2416-H110-055 

科資分類號: IB0002;SM0406 

關鍵詞: 虛擬商店 ;資料探勘 ;智慧型代理人 ;網路商店 ;電子商務 ;全球資訊網 ;倒傳遁神經網路

Virtual Store ;Data Mining ;Intelligent Agent ;Internet Store ;Electronic Commerce ;World Wide Web (Www) ;Backpropagation Neural Network 

中文摘要: 目前網路商店的運作模式,急需一套類似店員的智慧型機制,本專案運用資料探採與智慧代理人的技術,針對網路商店發展出一套令人滿意的虛擬店員系統,以提供資訊推薦的服務。此系統使用關聯分析來作資訊的推薦並應用倒傳遞神經網路來維護已發現的規則。我們同時建構了一個在網際網路上評估使用者資訊滿意度的衡量工具。實驗結果顯示,此系統能顯著地提升使用者滿意度。本研究所提出的增長式探採的方法,可有效地解決知識衝突與整合的問題,而新的使用者資訊滿意度的衡量工具也具有高度的效度與信度。這些方法與技術都能夠實際應用在搜尋引擎及電子商店上。 

英文摘要: Nowadays, the operating models of electronic stores very need an intelligent mechanism that likes a store clerk. This project employs data mining and intelligent agent techniques to develop a satisfactory virtual clerk system in order to recommend information on electronic stores. This system recommends information by using association analysis and applies backpropagation neural network to maintain the discovered rules. We also construct a measuring instrument to evaluate user information satisfaction (UIS) on the Internet. The experiment results reveal that the system increases user satisfaction significantly. The incremental mining method proposed by this research is efficacious to tackle knowledge contradiction and integration problems and the new UIS measuring instrument has high validity and reliability. These methods and techniques all can be applied to search engines and electronic stores actually. 

 

科資中心編號: RB8910-0390 

題名: 電子商務自動談判系統之開發

Automated Negotiation System for the Electronic Commerce 

研究者: 王俊程

Wang, Jiunn-Cherng 

機構名稱: 中正大學資訊管理研究所(CHCUINMG) 

經費來源: 行政院國家科學委員會(NSC) 

補助金額: 新台幣 575 仟元 

補助編號: NSC88-2416-H194-026 

研究日期: 1998/08 - 1999/07  

原件提供: 行政院國家科學委員會科學技術資料中心 NSC88-2416-H194-026 

科資分類號: IB0002 

關鍵詞: 自動談判 ;智慧型代理人 ;電子商務 ;遺傳演算法 ;談判支援系統 ;決策支援系統

Automated Negotiation ;Intelligent Agent ;Electronic Commerce ;Genetic Algorithm ;Negotiation Support System ;Decision Support System (Dss) 

中文摘要: 在多議題談判中,人們常常未能達到較佳的談判協議而使得談判結果沒有效率。跟傳統的談判活動比較起來,電子商務需要一個與傳統環境不同的談判形式。由於每個談判者對於不同的談判問題都有不同的判斷及認知,因此會有不同的效用函數,亦即談判問題結構。如果要達到總效用最大化的談判協議,談判者須要在不同的談判問題結構下,採用適合的談判策略才能達到。本研究之動機乃在探索電子商務談判中,導入談判策略知識的可行性。本研究探討智慧型代理人之自動化談判系統,並引入談判策略以導引談判的進行。整個談判過程使用遺傳運算法則,並且探討談判策略的導入,在不同的談判問題結構下所帶來的影響。同時也提出一個使代理人能從談判經驗中學習的方式,並研究談判經驗的影響。本研究使用系統模擬的方式,研究智慧型代理人在不同的談判問題結構下,使用不同談判策略及談判經驗所得到的結果。研究結果顯示出談判策略的導入,不管對於談判協議之個別效用或是總效用來說,都有顯著的影響。另外當代理人學習一段時間後,導入談判策略與無談判策略代理人之談判結果並無顯著差異。對於不同的使用談判問題結構,使用正確的談判策略對於談判結果也有影響。但是對於使用談判經驗的結果並沒有很顯著的差異。 

英文摘要: A well-established body of research consistently shows that people involved in multiple-issue negotiation frequently select pereto-inferior agreements that "leave money on the table." Compared with traditional negotiation activities, Electronic Commerce requires a different environment for negotiation. In addition, for certain structures of negotiation problems, maximized joint payoff settlements may be reached by wisely selecting proper negotiation strategies. In this research, we explore the feasibility of adopting negotiation strategies in Electronic Commerce environment. This research investigates the automated negotiation system that consists of simple, bounded rational, artificial intelligent agents using Genetic Algorithm(GA), an evolutionary computation approach. We construct a system to demonstrate how these agents use different negotiation strategies, simulate the environment of two-agent negotiation and investigate the effects of different strategies in different negotiation problem structures. Meanwhile, we present a way to adopt agent's experiences and study the effects of learning from negotiation experiences. The result shows that the adoption of negotiation strategies has significant impacts on both individual and total payoffs, despite the structures of negotiation problems. However, the effects of adopting strategies are less prominent as time elapses, while the GA approaches evolves to its best solution. We also find that the result of adoption of negotiation experiences is not significant. Finally, the result reveals that for certain problem structures, some strategies are more effective than others. 

 

科資中心編號: RB8910-0152 

題名: 智慧型助理在全球分散式物件網路上之研究

A Study on Intelligent Agents toward the Object Web 

研究者: 鍾興臺

Jong, Shing-Tair 

機構名稱: 淡江大學資訊工程系(TKUEIFE) 

經費來源: 行政院國家科學委員會(NSC) 

補助金額: 新台幣 459 仟元 

補助編號: NSC88-2213-E032-017 

研究日期: 1998/08 - 1999/07 

原件提供: 行政院國家科學委員會科學技術資料中心 NSC88-2213-E032-017 

科資分類號: IA0202;IA0407 

關鍵詞: 物件網路 ;智慧型代理人 ;全球資訊網 ;共通物件策託經紀人架構 ;爪哇 applet

Object Web ;Intelligent Agent ;World Wide Web (Www) ;Common Object Request Broker Architecture (Corba) ;Java Applet 

中文摘要: 由於全球資訊網路的普及,網路上的資訊包羅萬象,如何從這龐大的資訊網中有效找到使用者所關心的資訊就越形重要。目前,使用者端取得服務者端的服務,基本上是透過首頁中的表單或Java Applet的方式。如果資訊的來源是分散的,只以上述這兩種方法來做資訊整合的工作,明顯是不足的。使用表單來擷取資訊,常需從所傳回的HTML文件中過濾而得;而Java Applet則有安全性上的限制。為了改善上述服務者端提供服務的問題,及因應未來的全球分散式物件網路(Object Web),我們採用了智慧型助理(IA)與CORBA的技術,提出服務型助理,以提供使用者更便捷、更有效率的資訊擷取。服務型助理在架構上可依工作性質不同分成四層不同的助理層---通訊介面助理,控制助理,工作助理及資料庫助理。此架構不僅能與現行的WWW相容,且支援CORBA、KQML與Email的介面,使得使用者對於資訊擷取有更多的選擇。實際應用上,我們選擇以資訊變異快速且分散各處的股市資訊為開發的標的。 

英文摘要: WWW has become more and more popular in recent years. With huge information generated and scattered all over the Web, it becomes a challenge for users to retrieve the relevant information in a more effective and efficient manner. Currently, the interface to obtain the services in Web is through either the use of form or Java applet inside the homepage. But, both are insufficient for handling the information fuse once the information sources are scattered in different places. In this regard, using the form creates the need for information filtering because of the messages returned by the web server are mostly HTML documents; while using Java applet raises issues for security. To improve services given by service provider, we propose the Service Agents for fast, efficient information collection. This is by adopting technologies of intelligent agents and CORBA that aims to integrate the current Web with the Object Web. The Service Agents can be classified into four layers of agents consisting of communication interface agents, control agents, task agents and database agents. The different agents cooperate together. In the Service Agents architecture, the current WWW architecture and compatibility issues will be maintained. Service Agents also provide several interfaces including CORBA, KQML, and Email interfaces for users to better collect information. In practice, we target on the stock market as the related information is fast changing and scattered in different places. 

 

科資中心編號: RB8910-0136 

題名: 使用智慧型代理員於雙向無線PDA之推論策略與形式

The Strategies and Methods of Applying Intelligent Agent on Two-Way Wireless PDA 

研究者: 劉艾華

Liou, Ay-Hwa Andy 

機構名稱: 淡江大學資訊管理系(TKUMIFM) 

經費來源: 行政院國家科學委員會(NSC) 

補助金額: 新台幣 276 仟元 

補助編號: NSC88-2213-E032-015 

研究日期: 1998/08 - 1999/07 

原件提供: 行政院國家科學委員會科學技術資料中心 NSC88-2213-E032-015 

科資分類號: IA0202;IA0407 

關鍵詞: 個人數位助理 ;智慧型代理人 ;無線通訊 ;雙向傳呼 ;推論規則

Personal Digital Assistant (Pad) ;Intelligent Agent ;Wireless Communication ;Two Way Paging ;Inference Rule 

中文摘要: 本研究將傳統智慧型代理員的觀念應用於無線通訊環境中,就推論的策略與形式發展出一種機制-斥候代理員(Pilot Agent):將推論的實際動作交由伺服器端執行,代理員在客戶端僅使用少量資源進行使用者習性觀察、資訊的傳輸和遞送、以及將推論結果適當的表現出來。此機制既具備個人電腦強大又有效率的運算及推理能力,又可適時輔助一般使用者運用行動資訊遂行其需求,以達成隨時掌握資訊的目地。 

英文摘要: This research applied the concept of intelligent agents to wireless communication environment, and developed a scheme of strategies and patterns of reasoning-Pilot Agent. The reasoning actions of these agents were actually performed by inference server, while the agent at the client-end merely processes the observation of users' aptitude and transmission of information. The agent also presents the inference results properly by using sparing memory of client-end device. This scheme preserves the powerful and effective capabilities of computing and reasoning of PC, and assists users to accomplish their demands with mobile information to achieve the purpose of dominating information anywhere, anytime. 

 

科資中心編號: RB8703-0243 

題名: 網際網路上以智慧代理人為基礎之顧客化行銷系統

Intelligent Agent Based Customized Marketing System on the Internet 

研究者: 賴香菊

Lai, Hsiang-Chu 

機構名稱: 中山大學資訊管理系(NCSNIFN) 

經費來源: 行政院國家科學委員會(NSC) 

補助金額: 新台幣 268 仟元 

補助編號: NSC86-2416-H110-011 

研究日期: 1996/08 - 1997/07 

原件提供: 行政院國家科學委員會科學技術資料中心 NSC86-2416-H110-011 

科資分類號: IB1101;IA0410 

關鍵詞: 智慧代理人 ;顧客化行銷 ;電子商務 ;網際網路行銷

Intelligent Agent ;Customized Marketing ;Electronic Commerce ;Internet Marketing 

中文摘要: 網路商店雖急遽增加,但大多只是靜態地顯示公司相關產品訊息或提供簡單的線上訂購功能,並不能達到顧客化行銷的效果。本計畫係以達到顧客化行銷為網站開發之目的,提出一個以智慧代理人為基礎的顧客化行銷服務系統架構,並以校內的網路書店為基礎建構了智慧代理人的雛形系統,以驗證該系統架構之可行性。 

英文摘要: Up to now, most of the Web sites provides information only. Some have more than that. They provide functions of online ordering and e-mail communications, etc. However, these are not enough if a Web site wants to achieve the goal of customized marketing. This research provides a system architecture of intelligent agent based customized marketing system on the Internet. It also verifies the proposed architecture's feasibility by prototyping some intelligent agents. 

 

科資中心編號: RN9012-0025 

題名: 以機器學習理論與主成分分析求解多重品質特性之最佳參數設計 

研究者: 徐志明

Hsu, Chih-Ming 

機構名稱: 明新技術學院企業管理系 

經費來源: 行政院國家科學委員會 

補助編號: NSC89-2213-E159-033 

研究日期: 2000/08 - 2001/07 

出版日期: 2001/08 

頁冊數: 5頁 

原件提供: 行政院國家科學委員會科學技術資料中心, NSC89-2213-E159-033 

科資分類號: IB0500 

關鍵詞: 機器學習 ;主成分分析 ;參數設計 ;多重品質特性問題 ;品質特性 ;田口方法 ;光纖耦合器 ;單窗寬頻光纖分光耦合器

Machine Learning ;Principal Component Analysis ;Parameter Design ;Multi-Response Problem ;Quality Characteristic ;Taguchi Method ;Fiber Coupler ;Single Window Broadband Tap Coupler 

中文摘要: 基於機器學習理論與主成分分析,本計畫提出解決多重品質特性參數設計問題的整合性程序。所提之程序以最佳化融合製程參數為例說明,藉以改善1 比99 單窗寬頻光纖分光耦合器之光學特性與可靠度,並實行於台灣某家光纖被動原件製造商。實行的結果說明了本計畫所提程序的可行性,而比較結果也顯示所提程序在解決多重品質特性問題上優於傳統的田口方法。 

英文摘要: This project proposes an integrated procedure based on machine learning algorithms and principal component analysis to resolve the parameter design problems with multiple responses. The proposed procedure is illustrated through optimizing the fused process parameters, created during the development of fused bi-conic taper couplers to improve the performance and reliability of the 1% (1/99) single-window broadband tap coupler, and is implemented on a Taiwanese manufacturer of fiber optic passive components. Implementation results demonstrate its practicability, and a comparison also reveals that the proposed procedure outperforms the traditional Taguchi method in resolving multi-response problems. 

 

 

科資中心編號: RF9104-0136 

題名: 複決策結果之歸納學習法

Multiple-decision Outcome Inductive Learning Method for Knowledge Acquisition 

研究者: 鄭炳強

Jeng, Bing-Chiang 

機構名稱: 中山大學資訊管理系 

經費來源: 行政院國家科學委員會 

補助金額: 新台幣 504 仟元 

補助編號: NSC89-2416-H110-088 

研究日期: 2000/08 - 2001/07 

出版日期: 2001/07 

頁冊數: 6頁 

原件提供: 行政院國家科學委員會科學技術資料中心, NSC89-2416-H110-088 

科資分類號: IB0205 

關鍵詞: 歸納學習 ;機器學習 ;複決策 ;專家系統 ;決策 ;ID3演算法

Inductive Learning ;Machine Learning ;Multiple Dicisions ;Expert System ;Decision Making ;Id3 Algorithm 

中文摘要: 由現存的案例中建立決策樹的歸納學習對自動化知識擷取是非常有用的。大部分現存於文獻中的方法只處理單一決策結果的問題。在本計畫中,我們將發展一個處理複決策結果的歸納學習法。我們將空間分割重疊的觀念引入一般的歸納學習法當中。此方法先將每一個單決策的決策樹轉換成同一個特徵空間的分割,再將這些空間分割重疊以產生一個對複決策結果的空間分割。我們也進行一連串的實驗來顯示此方法的效果。 

英文摘要: Inductive learning that creates a decision tree from a set of existing examples is shown useful for automated knowledge acquisition. Most of the existing methods however, handle only the problems with a single-decision outcome. In this paper, we present an inductive learning method for multiple-decision outcomes. It introduces an overlapped partition-space concept into a regular inductive learning process. The idea is to transform the decision tree for each decision variable into partitions on a feature space and overlaps them to create a partition for multiple-decision outcomes. Experiments are performed which show the effectiveness of the new method. 

 

 

科資中心編號: RB9102-0413 

題名: 智慧型基因演算法的研發與應用

The Research and Development of an Intelligent Genetic Algorithm 

研究者: 何信瑩

Ho, Shinn-Ying 

機構名稱: 逢甲大學資訊工程系 

經費來源: 行政院國家科學委員會 

補助金額: 新台幣 499 仟元 

補助編號: NSC89-2213-E035-047 

研究日期: 2000/08 - 2001/07 

出版日期: 2001/07 

頁冊數: 5頁 

原件提供: 行政院國家科學委員會科學技術資料中心, NSC89-2213-E035-047 

科資分類號: IA0407 

關鍵詞: 基因演算法 ;智慧型基因演算法 ;大量參數最佳化 ;旅行銷售員問題 ;機器學習

Genetic Algorithm ;Intelligent Genetic Algorithm ;Large Parameter Optimization ;Traveling Salesman Problem ;Machine Learning 

中文摘要: 基因演算法(Genetic algorithm, GA)已被證明在解決少量參數最佳化問題有很好的效能。由於GA 隨機運算在遇上大量參數最佳化問題時將面臨不易收斂至全域最佳解和收斂速度緩慢等缺失,造成GA 應用到屬於大量參數最佳化的實際工程問題時,常無法發揮其效用。本計劃將對GA的理論及優缺點加以分析整理,提出以系統化推理取代GA 的隨機機制,稱此為(Intelligent genetic algorithm,IGA),並將其運用在旅行銷售員問題、圖形分割問題、模糊分類器設計、人臉三維模型建構、影像分割問題、圖形近似問題與群播繞路問題上,由實驗結果證實,IGA 可在這些問題上獲得優於現有作法之效能。 

英文摘要: Genetic algorithm (GA) has received a great deal of attention regarding their potential (high performance) as optimization techniques for complex problems with small parameters, and plays an important role on the theory of machine learning. Although there are many successes, GA still fails to find global optimum and converges slowly on solving large parameter optimization problems (LPOPs). In this project, an intelligent genetic algorithm (IGA) will be presented, which adopts a systematical reasoning mechanism to replace and improve the stochastic mechanism of GA. IGA will be applied to traveling salesman, graph partition,fuzzy classifier, facial modeling, image segmentation, curve approximation, and multicast routing problems. Experimental results show the high performance of IGA. 

 

 

科資中心編號: RB9009-0192 

題名: 利用PAT樹做為半結構性HTML文件資訊自動擷取之研究

Semi-structured Information Extraction Applying Automatic Pattern Discovery 

研究者: 張嘉惠

Chang, Chia-Hui 

機構名稱: 中央大學資訊工程系 

經費來源: 行政院國家科學委員會 

補助金額: 新台幣 347 仟元 

補助編號: NSC89-2213-E008-051 

研究日期: 1999/11 - 2000/07 

出版日期: 2000/08 

頁冊數: 8頁 

原件提供: 行政院國家科學委員會科學技術資料中心, NSC89-2213-E008-051 

科資分類號: IA0407;IA0202 

關鍵詞: PAT樹 ;半結構性文件 ;資訊擷取 ;自動擷取 ;超文件標示語言

Pat Tree ;Semi-Structured Document ;Information Extraction ;Automatic Extraction ;Hypertext Markup Language (Html) 

中文摘要: Information extraction (IE) from semi-structured Web documents is a critical issue for information integration systems on the Internet. Previous work in wrapper induction aim to solve this problem by applying machine learning to automatically generate extractors. For example, WIEN, Stalker, Softmealy, etc. However, this approach still requires human intervention to provide training examples. Hence , the other track to information extraction tries to save human effort. For example, Embley et. al. and Chang et al. present different approaches to record boundary identification of a single Web pages without any training example. Embley's work relies on the intra-page structure constructed by HTML tags (the parse tree), while Chang's work is motivated by repeated patterns formed by multiple aligned records. This paper expands Chang's work to IE and discuss the issues when applying pattern discovery for record identification, including the encoding schemes of HTML and ranking criteria of patterns to extract record boundary. 

 

 

[科資中心編號]RA81-2397

[題    名]

結合人工智慧及作業研究技巧發展複式法則解 n/1/t排程問題

Integrating AI and OR Technology Solving n/1/t Scheduling Problem.

[研 究 者 ]

利德江

LI,DER-CHIANG

[機構名稱  ]成功大學工業管理科學系(NCKBIMS)

[經費來源  ]行政院國家科學委員會(NSC)

[補助金額  ]222000

[補助編號  ]NSC80-0415-E006-03

[研究開始日期]1991/02

[研究結束日期]1992/01

[出版日期  ]1992/01

[頁 冊 數 ]50

[語    文]中文

[建檔日期  ]1994/05

[科資分類號 ]IB;IB0407;IB0201

[關 鍵 詞 ]

人工智慧;排程;延遲;作業研究

 

Artificial intelligence;Scheduling;Tardiness;Operation research

 

[摘    要]

      對大部分的生產排程問題,傳統上多利用 組合性最佳化

的概念,找出最適當的安排方式 。這些最佳化方法之中,如線

性規劃、非線性 規劃、動態規劃等,常能獲得精確的解答,計

算 上則需要較長的時間。因為這些方法大都視所 欲求解之

問題空間(Problems space)為一整體空間, 而發展一個通用之演算

法以解決問題。這種對 難易問題一視同仁的方式,通常較不

具效率,同 時所建立的模式,可能無法圓滿地解決實際的 問

題。因此,本研究乃結合人工智慧中的歸納 學習法(Inductive

learning)與作業研究之最佳化技 巧,以單機排程及部分工作不

得延遲之問題為 例,發展一套較具彈性與效率的多重法則

(Multiple algorithms),以使得總延遲時間(Total tardiness)為最小。

 

      This project combines inductive learning techniques from Artificial

Intelligence with traditional operation research methodologies to solve

scheduling problems. Focused problem is sequencing a set of jobs with unequal

importance on a single processor in order to minimize total tardiness. The

motivation of this study arises from the recognition of that most of the

traditional optimization algorithms are usually computationally inefficient

while dealing with np-complete problems such as tardiness problems because

they treated problems contained in the problem space uniformly or solve them

with identical procedure. Thus results in the fact that one may waste time by

misusing an inappropriate algorithm to too easy/difficult problems. On account

of this, this paper concentrates on exploring the field of combining inductive

learning techniques from Artificial Intelligence with traditional operation

research methodologies to divide the problem space into a number of

sub-spaces, and along with it, a multiple-algorithm is developed matching with

it in order to increase the efficiency of the algorithms. Used methodologies

are (1) an algorithm for sequencing jobs to minimize total tardiness as the

basis of optimization techniques and (2) the Attribute-Value object

representation techniques as the tool to decompose the domain problem space.

 

[科資中心編號]RA8510-2144

[題    名]

以類神經網路為學習基礎之智慧型排程系統

An Intelligent Knowledge Based Scheduling System with Inductive

Learning Capability.

[研 究 者 ]

劉志明

LIU,CHIH-MING

[機構名稱  ]清華大學工業工程研究所(NTHUINEG)

[經費來源  ]行政院國家科學委員會(NSC)

[補助金額  ]206000

[補助編號  ]NSC83-0415-E007-005

[研究開始日期]1994/02

[研究結束日期]1995/07

[出版日期  ]1995/06

[頁 冊 數 ]93頁

[語    文]英文

[建檔日期  ]1997/05

[出版情形  ]研究報告,93頁,民國84年6月

Report, 93p., Jun. 1995

[原件提供  ]行政院國家科學委員會科學技術資料中心

NSC84-2213-E007-035

[科資分類號 ]IB0407

[關 鍵 詞 ]

排程;人工智慧;類神經網路;決策樹

 

Scheduling;Artificial intelligence;Artificial neural network;

Decision tree

 

[摘    要]

      排程是生產計畫控制的一部分,在整個製

造過程中扮演著重要的角色。隨著不確定性與

複雜度的增加,大多數的排程問題已經被證實

NP-complete問題,傳統的OR技術已無法解決。因

,迫切地需要一個新方法來處理這些複雜且

是動態的排程問題。

      人工智慧技術的成熟提供了一個選擇性的

方案來處理這些問題。在這份研究報告中,提

供了一個具有歸納學習能力的新方法(SAOSS)。

AI的觀點來看,SAOSS提供了多元演算法,此方法

可以處理很多種的排程問題,並且比其他方法

更具有智慧、彈性與實用性。

      SAOSS是一個有效的歸納學習方法(CID3),藉由

轉換相對的決策樹到一個自我生成的類神