本論文中，我們設計了一個圖像素材庫。此系統的設計目的，在於能自動的幫助使用者整理圖像，對圖像資源作適當的分類，藉此省下使用者的人力以及時間。

　　本系統是採用高階特徵（Text-based）對圖片加以分類。首先利用自助式建構法對使用者手工建立的類別關鍵字作關鍵字延伸，以增加每個類別關鍵字的質與量。再將圖像的關鍵字與類別延伸後的關鍵字之間做相似度計算。完成此步驟之後，會得到一個初步的圖像分類結果，最後再利用特徵鑑別公制對分類的圖像作訓練，ㄧ直到分類結果不在改變為止，即表示完成圖像自動分類的整個流程。

　　在實作中，我們利用自助式建構法做關鍵字延伸達到圖片分類的目的，有別於一般的字義式關鍵字延伸，自助式建構法可以做到聯想式的關鍵字延伸，利用這樣的關鍵字延伸機制，可以做到WordNet等無法做到的圖像分類效果。圖像分類訓練方面，採用的是特徵鑑別公制的概念，即是先將特徵依照屬性作分類，再依照特徵的分類結果，將圖像作重新的分類。藉由自助式建構法與特徵鑑別公制兩者間的運用，讓系統對圖像的分類效能更逼近使用者手工分類的結果。

In this paper, we have designed a database that can automatically classify images, for the purpose of sorting through a large number of images more conveniently and thus save manpower and resources.  

This database is characterized by high level features (text-based) to image classifying. Its features include: extending a keyword through bootstrapping construction. First of all bootstrapping construction method extended words that the user manually inputted, and then increased the value and number of classificatory keywords. The keywords and classificatory keywords after extension underwent similarity value calculations. Finishing this step results in an initial classifying for images, and the step is repeated until there are no more changes in the classifications. 

Whereas common ways of extending a keyword deal with its definition, bootstrapping construction allows expansion through associative extension. This type of keyword expansion mechanism is capable of classifying images in ways that WordNet cannot. Aside from using bootstrapping construction to expand keywords and to classify images, we have also added a discriminative feature metric to increase the precision and recall rates of image classifying to our standards.

目 錄
第一章 緒論	1
1.1 研究動機與目的	1
1.2 研究內容	5
1.3 研究內容大綱	7
第二章 背景知識與相關研究	8
2.1 資訊檢索模式	8
2.1.1 二元模式(Boolean Model)	8
2.1.2 向量空間模式(Vector Space Model)	10
2.1.3 機率模式(Probabilistic Model)	12
2.2 自助式建構法(Bootstrapping)	14
2.3 自動分類演算法	16
2.3.1 最近鄰居決定法則(KNN rule)	16
2.3.2 貝氏分類法(Bayes classifier)	18
2.3.3特徵鑑別公制(Discriminative Feature Metric)	19
第三章 系統架構	23
3.1 系統架構	23
3.2 圖像類別建構子系統	26
3.3 圖像初步分類子系統	30
3.4 圖像訓練子系統	32
第四章 實作與討論	36
4.1 語意式圖像檢索系統之實作環境如下：	36
4.2 實作說明與探討	37
4.2.1 自助式建構法關鍵字延伸的準確性	38
4.2.2 自助式建構法關鍵字延伸對圖像分類的影響	41
4.2.3 特徵鑑別公制圖像訓練系統對圖像分類的影響	44
第五章 結論與未來研究方向	47
5.1 結論	47
5.2 未來研究方向	49
參考文獻	51
附錄	55







圖目錄
圖2.1	　THE COSINE OF Ө IS ADOPTED AS  	11
圖2.2	　BOOTSTRAPPING的運作	15
圖2.3	　投票機制	22
圖3.1　圖像分類系統架構圖	23
圖3.2　類別資料庫欄位介紹	27
圖3.3　自助式建構法流程	27
圖3.4　圖像初步分類流程	31
圖3.5　圖像訓練流程	32
圖3.6　圖像分類概念	34
圖4.1　經過訓練之後所得到的圖片分類結果	44

表目錄

表4.1 自助式建構法的初步實驗數據	38
表4.2 關鍵字準確率與相關文章數對應圖	39
表4.3 使用降低門檻值來做修正	40
表4.4 使用改進的方法所做出來的關鍵字延伸	40
表4.5 不同門檻值所得到不同的分類數據	42
表4.6 關鍵字延伸與否的F-MEASURE圖表	43
表4.7 經過訓練的六個類別的圖片回收率	45
表4.8 經過訓練的六個類別的圖片準確率	45

參考文獻

[1] Yu-Tao Huang, Integrating Semantics and Low Level Feathers  for thw Design of Image Retrieval Systems, Master’s	 Degrees in Science and Engineering, 2005.

[2] R. Baeza-Yates and B. Ribeiro-Neto,“Moderm Information 
Retrieval”, Addison Wesley Longman, Inc, May 1999.

[3] Kwiatkowska, M., Norman, G., and Parker, D. PRISM 2.0: A tool for probabilistic model checking. In Proc. 1st International Conference on Quantitative Evaluation
of Systems (QEST’04), 2004. To appear.

[4] Q. Iqbal, J. Aggarwal，　＂CIRES: A System for 
Content-Based Retrieval in Digital Image Libraries＂，Proc. International Conference on Control, Automation,   Robotics and Vision, pp. 205–210, Singapore, Dec. 2002.

[5] R. Jones, A. McCallum, K. Nigam, and E. Riloff，＂ 
Bootstrapping for Text Learning Tasks＂， In IJCAI-99 Workshop on Text Mining: Foundations, Techniques and Applications, 52-63, 1999.

[6] M. Thelen and E. Riloff. 2002，　＂A Bootstrapping Method for Learning Semantic Lexicons Using Extraction Pa
ttern Contexts＂， In Proceedings of the 2002 Conference
on Empirical Methods in Natural Language Processing.

[7] X. Liu and Y. Gong， ＂Document clustering with cluster
refinement and model selection capabilities＂， In
Proceedings of ACM SIGIR 2002, Tampere, Finland, Aug. 2002.

[8] Christopher D. Manming, Hinrich Schutze(2000), 
“Foundations of Statistical Natural Language Processing”,3rd ed,pp232-249

[9] Christopher D. Manming, Hinrich Schutze(2000), 
“Foundations of Statistical Natural Language 
 Processing”,3rd ed, pp267-307

[10] Yu, C., Ooi, B.C., Tan, K.L., Jagadish, H.V，　＂Indexing the Distance: An Efficient Method　to KNN Processing＂， VLDB, 2001.

[11] Dumais, S., J. Platt, D. Heckerman and M.Sahami, Inductive learning algorithms and representations for text  
categorization,Proceedings of the 7th International 
Conference on Information and Knowledge Management,
 Bethesda, MD, pp. 148-155, 1998.

[12] J. Yan, L. Wenyin,H. Zhang, Y. Zhang, “Thesaurus-aided Approach for Image Browsing and Retrieval”, 
International Conference on Multimedia and Expo(ICME2001), Japan, 2001.

[13] Yung-Hsiao Lan, Building Semantic Indexing for Images 
Retrieval Systems, Master’s Degrees in Science and Engineering, 2004.

[14] Kim, K. I., K. Jung, S. H. Park and H. J. Kim, Support vector machines for texture classification,IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 24, pp.1542-1550,2002.

[15] Yen-Ting Chiang, The Designing of a Semantic Image 
Retrieval System, Master’s Degrees in Science and Engineering, 2002.

[16] Christopher D. Manming, Hinrich Schutze(2000), 
“Foundations of Statistical Natural Language 
 Processing”,3rd ed, pp539-543

[17] W. Niblack, R. Barber, and et al. “The QBIC project: Query images by content using color, texture and shape”, Proc. SPIE Storage and Retrieval for Image and Video Databases, Feb, 1994.