A Novel Texture Analysis Method Based on Graph Spectral Theory
Tao Zhang, Hong Wenxue, Wang Jinjia · 2009
As an active topic in pattern recognition, the graph spectral is applied in clustering and segmentation. But issues in the analysis to image, especially the texture image, could not been retrieved till now. In this paper, we present a novel texture analysis method, which introduces graph spectral theory into the field of texture image analysis. At first, the image is partitioned into several sub images by window method, and then the sub images are degraded to sub graphs by cross section imaging technology. The sub graphs will be regarded as vertexes and the similarities between them is regarded as the edges in a global graph, and the feature vectors are obtained from the graph by graph spectral theory. The CBIR experiments based on Brodatz test suit show that the precision of the method in this paper elevates 6.03% than that of gray-level co-occurrence matrix.