Research And Realization Of Image Retrieval System Based On Multiple Features

With the computer network technology and multimedia development, the traditional text-based retrieval methods gradually exposed serious problems. Therefore, content-based image technology was appeared. It directly retrieve the content of the image characteristics (color features, texture features, shape features). Content-based image retrieval has become the hot research field.In this paper, the low-level visual of content-based image retrieval is the main line, starting from the basic principles and framework of the key technologies of the image retrieval.First, descript the three color spaces as RGB, YUV, HSV, analyze choice of color space, discuss the color quantization, introduce the YUV and HSV color space histogram method to extract the color features. Also present to use72bin reasonable quantification and3x3blocks seting different weight values to improve model based on the HSV color space histogram algorithm.Second, introduce several common texture analysis methods, including the GLCM (Gray Level Co-occurrence Matrix) and gray-scale texture moments and edge direction histogram analysis method. Based on the traditional GLCM, the image feature vector to carry out the Gaussian normalized on the basis of extracting image features, thus improving the traditional gray level symbiotic comments algorithm to enhance accuracy. At the same time, it focuses on improving the edge direction histogram analysis method which using the Canny operator to preprocess the image to improve image retrieval results.Third, This paper used color histogram and GLCM (Gray Level Co-occurrence Matrix), respectively as the color and texture features. This will not only make full use of the color characteristics, and also take full advantage of the gray scale information of the image texture features, which avoid a characterization of the one-sidedness of the image. Such a retrieval method obtained a better search results. At last, In order to compare the above study of several characteristics extraction methods, this paper used the Visual Studio platform to design and realize the image retrieval system prototype which based on the low-level visual content of images. Then introduced the various modules and system retrieval process, analyze the function of each module, and summarizes the present work and propose further research direction.The main innovations of this paper can be summarized as follows:(1) Improve the image retrieval method based on HSV color features. This improved method is based on the HSV color model, using72bin reasonable quantification and3x3reasonable blocks with user feedback mechanism is set to different weight values to achieve.(2) Improve the image retrieval method based on edge direction texture features. This improved method is based on the original edge direction histogram texture feature extraction, present the Canny operator to preprocess image before extracting feature to improve the image retrieval results.