Research On The Image Retrieval System Based On The Meanshift And EMD

In recent years, the research of content based image retrieval is a focus. In 2004, Feng jing proposed a region-based image retrieval system, it segments image according to the difference of color, features of image consists of features of regions of the image, this method is closer to the feelings of human than the method of global feature. The paper discusses the research of the region-based image retrieval system. It contains four modules: the preprocess module, feature extract module, distance metrics module and feedback module. This paper do some research in the areas as follows:In the preprocess area: according to the flaws of the Meanshfit algorithm in segment settings, this paper proposed a cross area-based algorithm. This algorithm can avoid the mistake of the judging cross area; as the result of experiments, this paper discusses how to choose the parameter; and implement Meanshift algorithm and wavelet algorithm.In the feature extract settings: according to the flaws of the wavelet texture-texture movement, the paper proposed a high rank based texture. As the result of experiments, this algorithm can avoid the texture movement effectively.In the feedback settings: according to the flaw of the estimat error of the SVM classification, the paper proposed a VQ-based SVM algorithm. The paper uses the VQ weight as the SVM weight. As the result of experiments, this algorithm can avoid the estimat error effectively.