Hidden Markov Fields (HMF) have been widely used in various problems of image processing. In such models, the hidden process of interest is assumed to be a Markov field that must be estimated from an observable process . Classic HMFs have been recently extended to a very general model called "evidential pairwise Markov field" (EPMF). Extending its recent particular case able to deal with non-Gaussian noise, we propose an original variant able to deal with non-Gaussian and correlated noise. Experiments conducted on simulated and real data show the interest of the new approach in an unsupervised context