Multiresolution hidden Markov chain model and unsupervised image segmentation

Several approaches have been proposed in the last few years to handle the problem of multiresolution image segmentation. In a Bayesian framework, models using Markov fields have been highly effective. However the computational cost can be prohibitive. Markov tree models were therefore proposed. Although fast, these methods do not always give good results. In this article, we propose a new approach using a Markov chain built by transforming multiresolution images into one vectorial process via a Peano type scan, the Hilbert scan. We work in an unsupervised context in which parameter estimation is carried out by using a mixture distribution algorithm, the ICE algorithm. Experimental results, including classification of multiresolution synthetic images and SPOT images, are presented in this paper.