- 學位論文
植基於顯著影像特徵之影像檢索技術之研究
The Study on Image Retrieval Technology Based on Salient Image Features
若您是本文的作者，可授權文章由華藝線上圖書館中協助推廣。
摘要
近年來，隨著資訊科技的發展與進步，影像處理的需求日漸增大，越來越多的媒體資訊，以數位化的形式在日常生活中使用，現今，不管是要取得或產生數位影像都是非常容易的事，但是要在龐大的影像資料庫中快速又精確地找到符合需求的影像，是一個值得深入研究的議題。 一般來說，以內容為基礎的影像檢索(Content-Based Image Retrieval, CBIR)技術，主要由以下兩個步驟來進行影像檢索 : 首先針對每張影像做特徵提取的計算，並儲存在特徵資料庫中。第二步驟則是藉由提供查詢影像(Query Image)給系統，經過相同的特徵計算並與資料庫中的每一張影像特徵相比較，找出最相似的影像回傳給使用者。 本研究提出一個顯著影像特徵檢索技術SMFD(Salient Point and Multi- Features Descriptor Based Image Retrieval Method)，首先透過哈里斯角點偵測(Harris Corner Detector)進行特徵點偵測以找出影像的特徵區域，再根據特徵區域中的彩度、明度計算其顏色平均值、變異數、顏色出現的次數及顏色聚集程度等特徵向量值，來抵抗影像中存在的幾何攻擊，透過特徵區域的匹配，計算其相似度並將最相似之影像回傳給使用者。在實驗結果，已測試不同類型之動物資料庫和Caltech 101人臉資料庫，對於影像檢索的精確率和召回率都有不錯的實驗結果，因此本方法可有效檢索相似之影像。
並列摘要
In recent years, with the development of information technology, There have an increasing demand for image processing. More and more media information with digital form is used in our everyday life. Now, to acquire or to generate digital images are very easy, but how to quickly and accurately meet the requirements to find the image in large image databases, is a intensive topic for further studies. In general, content-based image retrieval technology mainly includes the following two steps for image retrieval: first is the extracting features for each image and to store them in the feature database. The second step is to provide query by image to the system; to calculate similarity by the characteristics with the database features for each image and to identify the most similar images which are return back to the user. This research presents a significant image feature retrieval technology (Salient Point and Multi-features Descriptor based Image Retrieval Method, SMFD), The feature points are first detected by Harris corner to identify the features of image, and then the characteristics of the same color are formed to resist geometric attacks. The characteristics are the saturation, brightness, the average and, variance, the number of occurrences and the degree of the same color gathered. By matching feature regions through calculating the similarity, the similar images are return to the user. In the experimental results, different types of animals and the Caltech 101 face databases have been tested. As a result, the image retrieval precision and recall results are good, so this method can retrieve similar images of a image databases effectively.
參考文獻
[3]Chang N.-S., and Fu K.-S., “Query-by-pictorial-example,” IEEE Trans. Software Eng, SE 6 (6), pp. 519-524, 1980.
[4]Bezdek, J. C., “Pattern recognition with fuzzy objective function Algorithms,” Plenum, New York, 1981.
[5]Brunelli, R., and Mich, O., “Image retrieval by examples,” IEEE Transactions on Multimedia, vol. 2, no. 3, pp. 164-171, 2000.
[6]Lehmann T. M., Guld M. O., Deselaers T., Keysers D., Schubert H., Spitzer K., Ney H. and Wein B. B., “Automatic categorization of medical image for content-based retrival and data mining,” Compute Medical Imageing and Graphics, pp. 143-155, 2005.
[7]Rui, Y., Huang, T. S., and Chang, S. F., “Image retrieval: current technirues, promising directions and open issues,” Journal of Visual Communication and Image Representation, vol. 10, pp.39-62, 1999.