Image segmentation using the double Markov random field, with application to land use estimation

We describe the double Markov random field, a natural hierarchical model for a Bayesian approach to model-based textured image segmentation. The model is difficult to implement, even using Markov chain Monte Carlo (MCMC) methods, so we describe an approximation that is computationally feasible. This is applied to a satellite image. We emphasise the valuable additional information about uncertainties in the segmentation that can be gained from the use of MCMC.