The Research Of Color-based Image Retrieval

As the standard of people's living rise, the material culture requirements of people are increasing. Drab text already cannot satisfy the demand of human information, image as an informative performance media turn into people's perspective. With the high-technology such as digital technology and network rapid develop. The number of image is exponential growth, and applied to all kinds of society. How fast and accurately find the desired image from the image ocean has become the most efficient use of the bottleneck of the restricting image. The Content-based image retrieval technique attempts to break the bottleneck, gradually become the current research hotspot for domestic and foreign. And color is the most important visual image has been widely applied to the content-based image retrieval method. The main jobs of the dissertation are summarized as follows:The dissertation first reviewed the content-based image theory, especially the theory, technology, research status, selection of color, the color of quantitative representation, color feature extraction and similarity measure of the image retrieval are summarized.Global color histogram widely as the color features expression methods for it has the characteristics of rotation invariant. But it easy to appear the phenomenon of the different images have the same histogram, at the same time, some image histogram may exist large pixel value is zero color. These are global histogram defects, and significantly reduce the image retrieval efficiency. Therefore, this paper studies the effect of global color histogram and accumulates histogram retrieval. For some image can not extract the accumulate histogram, this paper studies the local accumulative histogram. And in view of the shortcomings of traditional local accumulative histogram once divided ignore the boundary color similarity, this paper apply the two dividing and weighted summation to obtain a new local accumulative histogram method. Through the experimental, the results of this method are feasibility and practicability. According to human vision habits, a new color quantitative method is devised in HSV color space. The method has abandon the color for brightness slam off and the low color saturation, make the quantification of color reduce 31 species, greatly reducing the number of colors and the computational complexity.A new retrieval method based on non-uniform block is summarized. According to the habit of human filming, the images are divided into five pieces, and then the center blocks are overlap blocked, to fully reflect the main part of the image,then combining with the new design of color quantitative methods. Through the comparative experiment between this method and uniform block retrieval method, this method has a little improvement in image retrieval.Based on the above theory, this paper developed an image retrieval experimental edition system with matlab. And this platform has confirmed the feasibility and effectiveness of the above theory.