Image Retrieval Based On Color And Shape Features

The main techniques of CBIR are discussed. First, the related researches on CBIR are introduced. Then, the key technique in CBIR is introduced.During the study of color space selection and quantification, we choose HSV color space by comparing of different color space and features, and then we quantify the HSV color model reasonably, extract color histogram as the feature to retrieve. The traditional color histogram does not contain the spatial location information as well as traditional block without taking into account of the linkages between the sub-blocks, we proposes an image retrieval method based on color-space histogram, which designs rectangular overlapped sub- regions to get color histogram in each region. It is mainly based on the characteristics that human eyes usually focus on the middle region of an image. And it can highlight the main parts of the image color distribution features. By extracting the various sub-blocks of the main color as the color characteristics, it can greatly reduce the color feature extraction and matching process in the computational complexity.During the study of shape feature extraction and matching, through the research and analysis of Moment Invariants and Fourier Descriptors, according to their invariant features, a new approach is introduced. The basic idea is: first to use the Moment Invariants for the selection of different images, then determines the similar images after using the Fourier Descriptors, and finally sorts the graphics sequence, it can greatly improve the efficiency of image retrieval.Taking into account of using single feature is not enough to express image data, we combine both color and shape features for image retrieval, we have the color and shape features of a linear weighted summation, then the similarity of two images is obtained, the weighted value of the ratio between the size representing the importance of color feature and shape feature. Experimental results show that an integrated search feature is better than the individual characteristic.