Skip to Main Content
Several frequently used feature vectors and segmentation methods are investigated, and a novel method is proposed for segmenting fully polarimetric SAR images by starting from the statistical characteristic and the interaction between adjacent pixels. In order to use fully the statistical a priori knowledge of the data and the spatial relation of neighboring pixels, Wishart distribution is integrated with Markov random field (MRF), and then an iterative conditional modes (ICM) algorithm is used to implement a maximum a posteriori (MAP) estimation of pixel labels. Although ICM has good robustness and fast convergence rate, it is affected easily by initial conditions, so a Wishart-based ML is used to obtain the initial segmentation map, with the statistical a priori knowledge also exploited completely in the initial segmentation step. Using fully polarimetric SAR data, acquired by the NASA/JPL AIRSAR sensor, the new approach is compared with several frequently used methods. Better segmentation performance, as well as better connectivity, less isolated pixels and small regions, are observed.