Unsupervised Segmentation of Multilook Polarimetric Synthetic Aperture Radar Images

International audienceThis paper proposes a new unsupervised image segmentation method for multilook polarimetric synthetic aperture radar (PolSAR) data. The statistical model for the PolSAR data is considered as a finite mixture of non-Gaussian compound distributions considered as the product of two statistically independent random variates, speckle, and texture. With different texture distributions, the product model leads to various expressions of the compound distribution. The method uses a Markov random field (MRF) model for pixel class labels. The expectation-maximization/maximization of the posterior mar-ginals (EM/MPM) algorithm is used for the simultaneous estimation of texture and speckle parameters and for the segmentation of multilook PolSAR images. Simulated and real PolSAR data are shown to demonstrate the method. Index Terms-Expectation-maximization (EM) algorithm, Markov random field (MRF), maximum-likelihood (ML), max-imization of the posterior marginals (MPM), polarimetric synthetic aperture radar (PolSAR)