Quantized image patches co-occurrence matrix: a new statistical approach for texture classification using image patch exemplars

Abstract

The statistical distribution of image patch exemplars has been shown to be an effective approach to texture classification. In this paper, the joint distribution of pairs of patches for texture classification from single images is investigated. We developed a statistical method of examining texture that considers the spatial relationship of image patches, which is called the quantized patches co-occurrence matrix (QPCM). In our method, the images are first slipt into small image patches, and then the patches are quantized to the closest patch cluster centers (textons) which is learned form training images. By calculating how often pairs of patches with specific quantized values (texton labels) and in a specified spatial relationship occur in an image, we create the QPCM for images representation. Moreover, we developed a fusion framework for texture classification by fusing 4 QPCM functions with specified neighboring spatial relationship and 3 other statistical representations of image patches, which is called QPCM-SVM classifier. The effectiveness of the proposed texture classification methodology is demonstrated via an extensive consistent evaluation in standard benchmarks that clearly shows better performance against state-of-the-art statistical approach using image patch exemplars.