On the regularization of segmented satellite images

The processing of satellite images involves two stages: the classification (segmentation) of the satellite image and the regularization of the segmented image. Usually, the classification is performed through the maximum a posteriori method (MAP), while regularization can be achieved by several methods. The authors focused on the regularization stage and the methods investigated are: the MAP regularization for discrete Markov fields (MAP-Markov), the mathematical morphology (MM) and the method of the median filter (MF). The paper presents an experimental, comparison study of these three methods. The performances and the weaknesses of each regularization method, establish some optimal parameters for the MAP-Markov was assessed and several practical recommendations for image regularization were formulated.