A Markovian classification method for urban areas of high-resolution SAR images

Aim to solve classification problems of high-resolution SAR images of urban areas, we proposed a method combining G0 distribution and Markovian classification. The recently proposed parameter estimation approach based on Mellin transform has been proven an accurate and efficient method for statistical models including G0 distribution. Markovian classification technique, preserving the spatial context information, can obtain good classification results. During optimization process, the Modified Metropolis Dynamics (MMD) algorithm is chosen, which can give the same global solution as Simulated Annealing (SA) algorithm and more efficient simultaneously. Applying on real SAR data, experiments results verified the better modeling capability of G0 distribution, and the quality by the classification that is obtained by mixing the model and Markovian segmentation is high and enable us to distinguish building, forest and sea.