Comparison of Multi-scale Approaches for Extracting Image Descriptors from the Co-occurrence Matrix

One of the first methods for analyzing the texture of an image was proposed in 1979 by Haralick, who introduced the co-occurrence matrix for calculating a set of image statistics. In this paper we focus on novel texture descriptors extracted from the co-occurrence matrix. It is well known that scale is important information in texture analysis, since the same texture can be perceived as different patterns at distinct scales. In this work we present, compare and combine different strategies for extending the texture descriptors extracted from the co-occurrence matrix at multiple scales. The texture descriptors are used to train a support vector machine and some different fusion techniques are compared. Our results are validated using seven image classification problems (mainly medical image classification problems). Our results shown that we improve the performance of the standard approaches. The code for the approaches tested in this paper is available at: http://www.dei.unipd.it/wdyn/?IDsezione=3314&IDgruppo_pass=124&preview=.