Research And Implementation Of Image Retrieval Based On Color And Texture Feature

Since 1990s, content-based image retrieval has appeared and quickly become the research focus in the field of image information retrieval. Considering the current status and problems of the CBIR, the paper has researched on the following aspects:The paper researched several key techniques of CBIR, including image feature extraction, image indexing, relevance feedback technology, and evaluation of image retrieval.The paper analyzed several existing color-based image feature extraction algorithms systematically. On the basis of that, the paper compared the effectiveness of several typical color feature extraction algorithms. To overcome the problem that global color histogram can't catch the spatial distribution information of colors, an improved algorithm of maximum connected region area histogram algorithm was proposed. Adopting this algorithm, the spatial distribution feature of color is reflected indirectly, while rotation and translation invariance of traditional color histogram are kept. Experimental results showed that the retrieval efficiency of the proposed algorithm is better than that of traditional color histogram. In addition, the paper researched several existing texture-based image feature extraction algorithm thoroughly. By experiment, the paper compared and analyzed the retrieval performance of GLCM, Gabor wavelet transform and Tamura texture features. In order to describe the image texture features more accurately, the paper presented an improved texture feature extraction algorithm, which combined GLCM with Gabor wavelet texture. Experimental results showed that compared with using single texture feature, the proposed algorithm can significantly improve the retrieval recall and precision.Single visual feature can't describe the image content completely and loses partial information so the retrieval performance of the system is relatively low. In view of the above, the paper implemented a multi-feature algorithm with both color and texture feature for image retrieval. The experimental results showed that compared with single feature, retrieval with multi-feature accorded with people's visual feeling better and so had better retrieval results. Finally, a content-based image retrieval demo system was designed and implemented using Microsoft Visual Studio 2010, Matlab and SQL Server 2000. The demo system provided an experimental platform for related research.