Unsupervised image segmentation using Markov random field models

We present two unsupervised segmentation algorithms based on hierarchical M arkov random held models for segmenting both noisy images and textured imag es. Each algorithm finds the the most likely number of classes, their assoc iated model parameters and generates a corresponding segmentation of the im age into these classes. This is achieved according to the maximum a posteri ori criterion. To facilitate this, an MCMC algorithm is formulated to allow the direct sampling of all the above parameters from the posterior distrib ution of the image. To allow the number of classes to be sampled, a reversi ble jump is incorporated into the Markov Chain. Experimental results are pr esented showing rapid convergence of the algorithm to accurate solutions. ( C) 2000 Pattern Recognition Society. Published by Elsevier Science Ltd. All rights reserved.