以天真貝氏為基礎的 “Query by Examples”圖片檢索系統 -以淡水觀賞魚為例

從網際網路發展至今，圖形檢索(images retrieval)的技術一直是學者投入的重點之一，目前為止，基於內容的圖形檢索技術(CBIR content-based image retrieval)[4][22]已有相當的成就。但是，今如果我們取得一張具有主題性質[註1]的圖片但不知名稱為何，例如動物、植物或是其他類別的圖片，我們希望知道這張圖片的名稱為何、叫什麼名字時，該如何知道？目前為止，電腦所獲得的影像資訊與人類的圖形認知和語意(semantic)表達仍有相當大的落差，往往同一張圖片，不同的人解讀會產生不同的認知，電腦本身並不能判斷使用者的主觀認知及真正想要的主題內容為何，因此就產生了所謂的語意斷層(semantic gap)，這也是大部分的圖片檢索系統在檢索『主題圖片』時結果往往不盡人意的主因[29]。因此，我們希望發展一套能查詢主題圖片名稱的系統，藉由『Query by Examples』[5][8]的概念，融入使用者主觀語意到圖形檢索系統中，強調以使用者為導向，讓檢索系統與使用者之間產生互動，以縮短語意斷層的差距。本研究以淡水觀賞魚為例，我們將魚的各特徵部位拆解成examples(圖例)，包括背鰭、尾鰭、臀鰭等[10][28]，讓使用者比較手邊的圖片或是看過的魚類，經過點選後，系統再以『Naïve Bayesian』[2]演算法分析預測使用者所選的組合，最後給予使用者最有可能種類的排列結果，使用Naïve Bayesian演算法的原因是根據本實驗資料的特性，比起決策樹或是類神經網路來說，計算速度快，較有效率[12][27]。簡單的說，當使用者不知道某種魚的名稱為何時，可以使用本套系統，以找出魚的名稱，圖片以及相關介紹。本文中我們所使用的『QBE查詢模式配合Naïve Bayesian演算法』，除了淡水觀賞魚之外，也可以應用在其他主題領域中，如花類圖片搜尋系統，鳥類圖片搜尋系統等等。

關鍵字

天真貝氏；圖片檢索；語意； Query by Examples(QBE)

並列摘要

From Internet development until now, the technology of image retrieval always is one focus that the scholars invested, the CBIR (content-based image retrieval)[4][22] systems had great achievement in the general-domain pictures domain [Footnote 1]. However, if we only know the category of pictures, such as: animal, plant, etc, but we don’t know the specific name, how can we do? The technology in the present stage, the distance still is quite huge that between the information gained from the pictures by the technology of computer recognition and by subjective cognition of human. If it only depends on the recognized technology to retrieval, the computer could not distinguish what kind of each user's cognition belonged to, and what kind of picture’s subject is. Therefore the “Semantic Gap” is presented to describe the difference subject cognition, this is the main reason that the retrieval results of “subject” pictures were usually hard to be acceptable [29]. Therefore, we hope to develop a picture retrieval system to be able to embed the user’s subjective meaning, and to inquire the names of the subject pictures, we take advantage of the concept of “Query by Examples” [5][8] and emphasize that let the users lead the system and make the semantic interaction between the retrieval system and the user, to reduce the Semantic Gap, and to enhance the retrieval process quality. Finally, we used “Naïve Bayesian” algorithm [2] to analyze and forecast the user’s combination of choice, then we gave the user the final results of most possible categories in descending order. In this research we adopt “The Freshwater Aquarium Fish” as an example. When the user got a fish picture and he want to know whatever about this fish, he may use this system to inquire the correct name. The reason we used Naive Bayesian algorithm is: Comparing the Neural Network and the Decision Tree, the Naive Bayesian algorithm is more suitable, more quickly and the accuracy is also much higher[12][27]. Moreover, “the QBE inquiry mode coordinating Naïve Bayesian algorithm” we introduced in this research, besides the Freshwater Aquarium Fish domain, also may apply in other subject domains, for example: flowers pictures retrieval system, birds pictures retrieval system, butterfly pictures retrieval system and so on.

並列關鍵字

Image Retrieval ； Na ； Query by Examples(QBE) ； Semantic

參考文獻

2. Duda, R. O., and Hart, P. E., Pattern Classification and Scene Analysis, Wiley, New York, 1973.

4. Gudivada, V. N., and Raghavan, V. V., “Content-Based Image Retrieval Systems”, IEEE Computer, Vol.28, No.9, Sept. 1995, pp.18-22.

5. Gunsel, B., and Tekalp, A. M., “Shape similarity matching for query-by-example”, Pattern Recognit, Vol.31, 1998, pp.931-944.

6. Gupta, A., and Jain, R., “Visual Information Retrieval”, Communications of the ACM, Vol.40, No.5, May 1997, pp.70-79.

7. Han, J., Kamber, M., Data Mining: Concepts and Techniques, Morgan Kaufmann Publishers, 2001, pp.296-299.

國際替代計量

以天真貝氏為基礎的 “Query by Examples”圖片檢索系統 -以淡水觀賞魚為例

未授權

主題瀏覽