Content Based Image Retrieval of 林宏銘、傅家啟
E-mail: [email protected]
ABSTRACT
Micorcalcification is an early index of breast cancer. Computer Aid Diagnose (CAD) system can help users to diagnose
microcalcification clusters. When users have doubt on the output of a CAD system, it is not able to provide a second opinion. To solve this problem, we develop a content-based image retrieval (CBIR) system of mammographic microcalcification clusters. The goal of the proposed CBIR system is to provide a set of pathology verified images relevant to the query image as the reference for users. The two essential aspects of CBIR are the feature extraction and the definition of similarity. In this thesis, thirty-five features are extracted from each cluster. Since not all features are useful, the Sequential Forward Selection(SFS) technique is applied to select 6 most useful features for input feature vector of the CBIR system. In this thesis, we design investigate three types of similarity indices to evaluate the system performance. The first is to calculate the angle between the feature vectors extracted from the query image and the images in the database. The second is to calculate the Euclidean distance between the feature vectors extracted from the query image and the images in the database. The last, we applied a general regression neural network (GRNN) based learning algorithms. To provide an objective measurement for the desire output, we apply the angle between the feature vectors extracted from the query image and the images in the database as the desire output value for GRNN training. In addition, we formulate a linguistic query system, which allow users to query images by inputting features though semantic query. By applying the images from the Nijmegen Digital Mammogram Database, experimental results showed that the GRNN, with precision rate 0.9405 and recall rate 0.8095, out performs the similarity indices formulated by the angle and Euclidean distance.
Keywords : Content-Based Image Retrieval ; Microcalcification Clusters ; General Regression Neural Network ; Linguistic Query Table of Contents
目錄 封面內頁 簽名頁 授權書 iii 中文摘要 iv ABSTRACT v 誌謝 vii 目錄 viii 圖目錄 x 表目錄 xii 第一章 緒論 1 1.1研究目的 與動機 1 1.2研究範圍 1 1.3研究方法 2 第二章 文獻探討 3 2.1 前處理 3 2.2 特徵萃取 5 2.3 內容導向影像檢索 6 2.3.1 相似度 比對演算法 8 2.3.2 影像檢索於數位化乳房x光影像之文獻 9 2.4 語意式查詢系統 13 第三章 研究方法與架構 18 3.1 研究架構 18 3.2 前處理 19 3.3 特徵萃取 21 3.4 內容導向影像檢索系統 21 3.4.1 相似度計算 22 3.4.2 績效衡量 27 3.5 語意式查詢系統 28 第四章 實驗結果與分析 32 4.1前處理 32 4.2 特徵萃取 33 4.3 相似度演算法 34 4.3.1餘弦法 34 4.3.2 歐基里得距離法: 38 4.3.3 GRNN相似度計算法 42 4.4語意式查詢系統 46 第五章 結論與未來研究發展 51 5.1 結論 51 5.2未來研究發展 52 參考文 獻 53 附錄一 廣義神經網路 56 附錄二 Fuzzy C Mean 演算法 59
REFERENCES
1. 傅家啟、李三剛、溫嘉憲、蔡明倫、林宏銘，“乳房X光影像中微鈣化之強化、特徵萃取及辨識”，中華放射醫誌，2003。 2. 羅強 華, “類神經網路:MATLAB的應用”, 清蔚科技股份有限公司, 2001. 3. 蘇木春、張孝德,"機器學習 類神經網路、模糊系統及基因演算法 則”,全華科技圖書 4. Al B., "Handbook of image and video processing ",Academic Press,2000 5. Azevedo P.M. and Santos R.R. , Traina AMJ, Traina Jr. C, Bueno JM “Image Retrieval Based in Texture Content Aiding Breast Cancer Diagnosis”, IWDM , 2000. 6. Chiu C. Y., Lin H. C., and Yang S. N. , "Learning user preference in a personalized CBIR system" ,The 16th IAPR International Conference on Pattern Recognition, Quebec City, Canada, Aug. 11-15, 2002. 7. Chu, W., Sathiya S., Chong J.O. ,“A Unified Loss Function in Bayesian Framework for Support Vector Regression”,2001. 8. Colin C., "Kernel methods a survey of current techniques ",Neurocomputing,2000. 9. El Naqa, I., Yang Y., Galatsanous N. P.,“Image retrieval based on similarity learning,”IEEE Int. Conf. Image. Proc., Vancouver, Canada, 2000. 10. Haralick , R. M., Shanmugam, K., Dinstein, I., “Textural Features for Image Classification”, IEEE Trans. Systems, Man, and Cybernetics, Vol. SMC-3, no. 6, Nov. 1973. 11. Kim, J. K., Park, H. W., “Statistical textural features for detection of microcalcifications in digitized mammograms”, IEEE Trans. Medeical Imaging, Vol. 18, no. 3, Mar. pp. 231-238, 1999. 12. Lin H. C., Chiu C. Y., and Yang S. N., "LinStar texture: a fuzzy logic CBIR system for textures," The 9th ACM International Conference on Multimedia, pp. 499-503, Ottawa, Ontario, Canada, Sep. 30-Oct. 5, 2001. 13.
Liu, H., Motoda, H., “Feature Extraction, Construction and Selection: A Data Mining Perspective”, Kluwer Academic Publishers, 1998. 14.
Medasani S. and Krishnapuram R.,"A Fuzzy Approach to Complex Linguistic Query Based Image Retrieval",1999. 15. Medasani S. and Krishnapuram R.,”Image categorization for efficient retrieval using robust mixture decomposition.” 16. Rui, Y.and Huang T. S.,”Content
based image retrieval with relevance feedback in Mars”, image processing , Vol.2 ,pp 815-818, 1997. 17. Sonka M., Hlavac, V.and Boyle, R.,
“Image Processing, Analysis, and Machine Vision”, Brooks/Cole Publishing Company, 1999. 18. Vapnik, V., “The nature of statistical learning theory”, Springer Verlag, 1995. 19. Wong, S., “CBIR in medicine : still a long way to go ”,IEEE workshop on Content-Based image retrieval Access of image and video libraries, Santa Barbara CA, June 1998. 20. Yoo H.W., Jung S.H. and Jang D.S.,”Extraction of major object features using VQ clustering for content based image retrieval ”, Pattern Recognition 2002 21. http://www.doh.gov.tw/statistic/data/縣市癌症 與死因統計結果/91縣市順位.xls
