Selection of optimal features for texture characterization and perception

Different approach to texture characterization can be considered. In this work texture are analyzed through second order statistical measurements based on the Grey-Level Co-occurrence Matrix proposed by Haralick [1]. By this method is possible to compute 22 different features to describe texture. Usually, in previous works, only 5 features are considered among the complete set, but no reasons are exposed for that selection. In this work, using Principal Component Analysis, the set of features is studied and 5 features, different from former, are proposed as the most convenient describing and characterizing the considered textures. Finally, the relationship between the proposed features and perception of texture is analyzed.