The Research On CBIR And System Realization

With the rapid development of information industry, people can get more and more image information, both in living and working, for the information communication. That makes the image data increasingly rising dramatically. How to rapid and effective management these huge amounts of image data becomes a challenging problems. Content-based image retrieval (CBIR) is an effective technique for the image analysis and management. In recent years, CBIR is a very active research area and has been applied in more and more fields.In this paper, some of the key techniques in the CBIR have been introduced. It is mainly about the visual feature extraction, similarity matching, relevance feedback and the performance measurement of the CBIR. The features of the images include color, texture, shape and spatial feature and so on.The color feature is an important visual characteristic of CBIR, has been extensive used. Color histogram, color moment and so on are the color feature extraction methods.In the HSV color apace, quantified color histogram has been extracted. Using eigenvector absolute distance algorithm of image matching, an image retrieval system model has been realized. Then the shortcoming of traditional color histogram has been deeply discussed.The single feature of images can't fully express images content information, make that the precision of CBIR be limited. To overcome the short points, fractal texture information is used in the CBIR. The image color feature and fractal texture feature are comprehensive extracted. The computation of image fractal dimension has been analyzed. Make sure that DBC is a simple and accurate method for the image fractal dimension estimating.According to character of DBC method, we proposed quantized fractal dimension histogram. The image fractal dimension of the every color channel has been computed in the HSV color space. Make the approximate linear ratio relationship between the roughness of images and the Fractal Dimension. Experiment shows that fractal dimension quantification histogram, can find the same roughness of the images, and satisfied the psychology of human sensation about image roughness. The image features extracted in this paper both are in accord with the human visual perception of images information. The experiments show that the method of this paper is similar to the human visual sense and can get an effective search results, even can extract the abstract means of the images. The retrieval precision is higher than traditional color CBIR.