Research On Techniques Of Content-Based Image Retrieval At Digital Libraries

In the following of multimedia information technology and communication technology advancement and rapid development of Information Highway, the digital library has emerged, as the times require. Digital libraries have a lot of image resources, then there is an important theme how we can search these resources effectively. So the technology of content-based image retrieval (CBIR) has gradually become a present research hotspot.There are two critical issues for the technology of CBIR. One is how to describe the content of image. Another is how to measure the distance between two image's features. This paper do research surround with these two problems. Feature representation and extraction is the basis of CBIR. This paper provides acomprehensive survey for these low-level features from three aspects------color,texture and shape. For color feature, two color spaces in accord with person's perception are introduced. And we discuss some methods of color features extracted that are usually used. In this paper three methods of description for texture features have been discussed in detail. For shape feature, the thesis introduces several methods of edge detection to draw images in objects' outline, then on this foundation it puts forward a kind of fast edge detection algorithm, tests for the algorithm and makes some experiment results.This thesis applies the newest theory about statistical learning at present, support vector machine (SVM), in the process of image feedback retrieval. Then we bring forward the feedback processes using One-Class SVM and Two-Class SVM methods. In the basis of these we put forward a sort of improved technology. During the interactive procedure, the positive and negative sample images respect to the image marked by users both in current circulation and in the historical circulation are learned for constructing a SVM classifier, with which we classify the database images again. Experiments demonstrate that it can search more relevant images when there are limited training samples.Finally this paper uses programming tools to carry on the retrieval processes and uses the color and the texture features as the sole characteristics separately to confirm the results based on these algorithms. And a technique based on co-occurrence matrixis proposed. It realizes image retrieval through the method of edge detection algorithm to extract texture features of objects in images. In the retrieval process, it unifies the SVM method of relevance feedback, which this article improves, then produces the corresponding experimental results.