Unsupervised Change Detection in Sar Images Using a Multicomponent Hmc Model

In this work, we propose to use the Hidden Markov Chain (HMC) model for fully automatic change detection in a temporal set of Synthetic Aperture Radar (SAR) images. First, it is shown that this model can be used as an alternative to the Hidden Markov Random Field (HMRF) model in the image differencing context. We then propose a novel approach, called joint characterization, whose principle is to consider that the ‘before’ and ‘after’ images are a unique realization of a bi-dimensional process. Parameters estimation is performed from a multicomponent extension of the HMC model and thematic change can be detected according to the joint statistics of the classes in the images. Preliminary experiments show promising results.