Estimating the Gauss-Markov Random Field parameters for remote sensing image textures

Although there are some recent characterizations of Multivariate Gauss Markov-Random Field (MGMRF) models, these are limited to cases where the interaction matrix coefficients are modeled with some special form. We extend the modeling and parameter estimation for the interaction matrix coefficients for a general anisotropic MGMRF. Although the MGMRF is a natural generalization of its univariate counterpart, there are new problems which hold in the multivariate case but do not hold in the univariate case. The results show that in general, there is an improvement in the classification performance in the generalized model compared to competing MGMRF models.