Fast Segmentation of Markov Random Fields Corrupted by Correlated Noise

Markov Random fields (MRF) represent a powerful mathematical model and they are used in several areas, but it is almost impossible to perform exact analytical calculations when using MRF and we must use approximations and iterative methods that are greedy in terms of time and computing resources. In the literature, proposed MRF methods for Bayesian and parameters estimation are complicated for implementation and represent many disadvantages in practice. We propose in this work, and in order to remedy the problems mentioned above, a very simple MAP-MRF framework based mainly on local conditional probabilities, contrary to the existing solutions in literature where we rely on the energy function model. Two powerful models based on the proposed framework are then presented. They will be compared with two recent works to show how they are more efficient with respect to classical models when it comes to the unsupervised segmentation of corrupted data with correlated noise.