An efficient approach for texture-based image retrieval

The approach of co-occurrence matrix is widely used in texture-based image retrieval, yet it does not perform very well when the image gets a rotation and great change in size. In this paper, the spatial information of the image is applied to image retrieval based on generalized image co-occurrence matrix. Firstly, a smoothed image is obtained by using Sobel operator. Secondly, we construct a matrix with the correspond pixels in both of these two images which is called generalized image co-occurrence matrix here. Thirdly, a set of statistical features according to this matrix is obtained. Finally, a vector composed with this set of statistical features is regarded as the final characterization of the image and is kept in the feature database. Experimental results show that the retrieval efficiency is improved largely compared with the classic co-occurrence matrix method.