Unsupervised Bayesian classification of SAR-images

A nonsupervised method of Bayesian contextual radar image segmentation is presented. The hierarchical image model (HIM) from Kelly and Derin is adopted. In contrast to their global segmentation method, a local method is used in order to speed up the segmentation. The accuracy and the rapidity of the blind method and the contextual method are compared using different synthesized images. The results show that the blind method converges much faster, but the contextual method is much more accurate. These two method are combined using the blind method as a preprocessing (presegmentation) for the contextual method. The moment to switch between the two methods is implemented in such a way that the final segmentation is found as quickly as possible. The contextual SEM is shown to be a robust method with regard to the variability of spatial dependence.