Region-based Approaches and Descriptors Extracted from the Co- Occurrence Matrix

Recently proposed texture descriptors extracted from the co-occurrence matrix across several datasets is surveyed and validated in this paper; moreover, two new methods for extracting features from the Gray Level Co-occurrence Matrix (GLCM) are proposed. The descriptors are extracted not only from the entire GLCM but also from subwindows. These texture descriptors are used to train a support vector machine. We also explore region-based approaches, which use different methods to divide each image into two different regions; different descriptors are extracted from each region. In this work methods based on saliency detection, edge detection, and wavelets are compared, and some of their fusions are reported as well. Region-based approaches are combined with different methods for extracting features from the GLCM and with three state-of-the-art descriptors: local ternary patterns, local phase quantization, and rotation invariant co-occurrence among adjacent local binary patterns. Experimental results show that the tested approaches improve performance of standard methods. The generality of the proposed descriptors is demonstrated on 15 datasets, and different statistical comparisons based on the Wilcoxon signed rank test are reported that confirm the goodness of the proposed approaches. Experiments show that the new methods for extracting features from the GLCM greatly improve the standard features that are typically extracted, and that the regionbased approach boosts the performance of texture descriptors extracted from the whole image. The MATLAB source code of all the proposed approaches will be made available to the public at https://www.dei.unipd.it/node/2357.