Research On Methods Of Vision-Based Image Retrieval And System Implementation

Under the rapid development of the multimedia, network, and especially Internet technology, information continues to expand. People can access to a large number of image data frequently. The traditional keyword-based retrieval method has been unable to adapt to the needs of the image retrieval, how to organize, manage, query and search the large-scale information has been an important topic before us. Therefore, the Content-based Image Retrieval (CBIR) technique emerged. And it has gradually become a hot research topic.This article firstly introduces the related research on CBIR both at home and abroad and some problems and difficulties encountered in the research. Then, after analyzing the system architecture and key technologies of CBIR, the paper discusses the process and algorithms of image retrieval based on color and texture. As for color-based image retrieval, the paper introduces RGB, HSV and YUV color spaces and conversion between them. And then this paper proposes a method of based on an improved color histogram. This method chooses the HSV color model which is in accordance with the perception of human eyes, segments the image to some sub-blocks and makes the main parts of image or the interested sub-block of image increase in weight. These will deal with all the shortcomings of the global color histogram which includes no spatial information and the conventional distribution which pays no attention to the main parts or the interested areas. Also, these will greatly enhance the intelligence of retrieval and the rate of recall and precision. As for texture-based image retrieval, this paper focuses on introducing a feature extracted method of based on gray level co-occurrence matrix of sub-block. The method uses for YUV texture space model. And due to the large calculation of gray level co-occurrence matrix, the method compresses the gray-scale of image. And in order to highlight the central region of image, the method segments the image into several sub-blocks, and then calculates the co-occurrence matrix and texture feature of every sub-block. This method improves the efficiency of image retrieval.Based on the above methods, this paper sets up an experimental platform of Content-based Image Retrieval and achieves a variety of retrieval methods. Through this experimental platform, the paper analyzes the experimental results which are obtained by the methods based on color, texture and the combination of these two features. The Analysis show that method based on the combination of color and texture is better than the method based on single feature.Finally, the paper summarizes the research and outlooks the future research work.