Spatially non-homogeneous potts model parameter estimation on higher-order neighborhood systems by maximum pseudo-likelihood

This paper addresses the problem of maximum pseudo-likelihood estimation of the non-homogeneous Potts image model parameters using higher-order non-causal neighborhood systems in a computationally efficient way. The motivation is the development of a new methodology for contextual classification that uses combination of sub-optimal MRF algorithms for multispectral image classification, which requires accurate parameters estimation. Our objective is to make multispectral image contextual classification fully operational without human intervention. The results show that the method is consistent with real data and in the presence of random noise.