The Markov random fields in functional neighbors as a texture model: applications in texture classification

The main objective of this work is to design an approach for the study of textures that is capable of handling textures of different sizes in the same resolution scale. In addition, we want this approach to be independent from the images it analyzes in order to make it valid for the largest possible number of application fields. These considerations have led us to using a Markov random field model in which we have modified its probabilistic dependence so that it is capable of analyzing microtextures and macrotextures simultaneously. These modifications are carried out by means of the introduction of a new non-standard system of neighbors, called functional neighbors. Finally, we show how Markov's random field with a system of functional neighbors provides better results in texture classification tasks than with a system of physical neighbors.