Bayesian segmentation of images: Comparison of contextual and global methods

The Bayesian segmentation methods are investigated in considering the spatial structure of multidimensional pictures. Depending on whether one pixel is segmented from a neighbourhood-based local context or from the entire picture, the methods are classified as contextual or global methods. In the two first methodologic sections, general principles of segmentation algorithms and their close link with spatial modelling have been analysed. A bibliographic synthesis is given. Finally, contextual and global methods are compared with experiments in the third section and the influence of regularization parameters is examined.