Permutation entropy: Texture characterization in images

Pattern recognition in time series and in image textures are meaningful areas of information processing. The choice of appropriate descriptors is a fundamental step for the later pattern recognition, considering the degree of discrimination provided by the different possible descriptors. The Permutation Entropy (PE) descriptor is known in the literature because it allows to determine the complexity in time series based on the comparison of neighbor values. It presents great advantages in comparison with other descriptors: speed, robustness, straightforward computation and invariance with respect to nonlinear transformations. In this work, it is proposed the application of different techniques of PE in the characterization of textures in images. The experiments were performed considering both synthetic and real texture images. Different approaches to PE were used to create new descriptors and the discriminating ability of each of these descriptors in the recognition of different textures was analyzed.