Fuzzy co-occurrence matrix fusion based texture feature extraction

In this study, a novel fuzzy co-occurrence matrix to extract texture feature is presented. The impact of directional difference is eliminated through multi-angle fusion of gray level co-occurrence matrix (GLCM). Fuzzy c-means is introduced into gray level co-occurrence matrix, and the membership of each pixel to texture unit is calculated. Then second-order statistics, such as energy, entropy, contrast, homogeneity and correlation, are deduced to describe texture characteristics. Experiments assisted by one-against-rest support vector machine on benchmark texture datasets have shown that the proposed method, considering the directional differences and uncertainty probability of GLCM, provides a better rotation invariance and robustness than the other improved GLCM.