unsupervised classification of polarimetric sar image using scattering mechanism and markov random fields

in this paper, an unsupervised classification method using spatial contextual information for polarimetric sar (polsar) image classification is proposed. first, an unsupervised classification based on 2d h/▁α plane was performed, using cloude/pottier target decomposition algorithm. in order to compute the initial values of the cluster centers and hence a rapid convergence of the algorithm, the output of the h/alpha classification have been considered. then, using discriminant function derived from bayes theory, the classification based on map criteria was carried out. in the map criteria, the wishart distribution was used as the distribution of the polsar data. we also used markov random field algorithm for modeling the spatial information to calculate prior probability of classes.in order to enhance the classes separation, two image data from two different seasons, were used. in this study, the radarsat-2 satellite images in a forested area known to petawawa, canada, were used.