Key Technology Research, Content-based Image Retrieval

The rapid development of computer, multimedia and Interact techniques has produced too large amount of images .Therefore, it becomes an urgent problem that how to fred needed image efficiently in large-scale image database. One effective ways has been proposed to solve the problem: image retrieval based on content. The main search content about the technique include: extracting feature, relevance feedback and index technology.in this paper, we make research in each step of it.Our research work focus on the following aspect:(1) To improve image retrieval speed and accuracy, some noval feature are presented to represent image low-level feature,including the mayor color, global color and shape feature of image using binary information;to compute seminary measure of two image,we present a adaptive weighted color histogram ,and introduce concavity tree to represent shape feature.(2) A noval image indexing method, indexing tree based on SOON clusting and semilar clusting, is presented, which is two level indexing: SOON clusting indexing and semilary clusting tree indexing. SOON clusting method is better performance than others clust ,so it is used to image clusting and we present a indexing based on SOON,on the base ,we present indexing tree based on seminary clusting tree.(3) We present three novel relevance feedback method based on machine learning. First, to void unbalance of positive and negative sample we use relevance feedback based on BSVM methods; then in order to improve the accurate rate, we present relevance feedback based on BOOSTING and BSVM;on the base ,in order to improve the speed of image retrieval ,we present the long learning relevance feedback algorithm.(4) We design an integrative image retrieval system, which include all our presented method above, and design all code.