Texture Classification via a New Statistical Feature Extraction from Co-occurrence Matrix

In this paper, a statistical approach based on auto-covariance function is presented. Some statistical features are obtained from co-occurrence matrix. Two steps are performed to classify images. Initially, the co-occurrence matrix is obtained from images and then, auto-covariance function is applied to co-occurrence matrix in order to extract the proposed feature. The linear discriminant analysis (LDA) classifier is derived to classify the extracted features. The classification results gains better performance by achieving the benefit of proposed feature than using the former co-occurrence features. It should be noted that classification is independent of training data. The performance of the proposed method is evaluated by using CUReT data base.