Research Of Image Retrieval System Based On MPEG-7

With the rapid development of multimedia and network technology, content based image retrieval CBIR has become one of the hot spots in the multimedia information retrieval research.The main idea based on image retrieval contained in color, texture, shape and spatial relationships of objects as low-level image features and semantic features of high-level information to analyze the images, MPEG-7 formally regulate the description of multimedia content and the international standards, it describe the relationship between content and information,make the achievement of rapid and efficient image retrieval possible.This dissertation mainly focuses on MPEG-7 color, texture, shape and the high-level semantic features to analyze the image, combined with relevance feedback image retrieval techniques and analysis the image retrieval, Major research has the following points:Firstly,According to MPEG-7 standard put forward MPEG-7 domain color and color layout to extract the color descriptors; Gabor wavelet and edge histogram to extract the texture descriptors; Use ART and CSS to extract the MPEG-7 region and contour shape descriptors; Dynamically adjust the weight of multiply features to realize the multiply features of image retrieval; through the bottom of MPEG-7 descriptors to analysis image retrieval experiments show that adjust the weight of multiply characteristics is the most efficiency image retrieval method.Secondly,Through support vector machine SVM feedback technology, proposed incremental SVM learning algorithm. Combined with color,texture, shape and multiply characteristics of images to analysis SVM relevance feedback, and the experiment also compare the incremental SVM learning related algorithm. The experiments show that gradual incremental SVM learning algorithm improve the efficiency of image retrieval best.Thirdly,According to MPEG-7 Image Description Definition Language DLL and XML language to extract MPEG-7 image semantic, and then combined with XML semantic classification of training to realize the MPEG-7 semantic image retrieval.This image retrieval way can improve retrieval efficiency better than traditional keyword retrieval.