A comparison of clustering fully polarimetric SAR images using SEM algorithm and G<sup>0</sup><inf>P</inf> mixture modelwith different initializations

Abstract

This paper presents a comparison between two types of initializations for multilook polarimetric SAR image segmentation: a random partition and a sample quantile partition. These are the inputs of a stochastic expectation-maximization algorithm that uses a mixture of G <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">0</sup> <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">P</sub> distributions to describe the data. The parameters are unknown, and estimated by the moments method. The G <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">0</sup> <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">P</sub> law is able to describe different type of targets, like urban areas, vegetation and pasture. The experimental results on real PolSAR data are reported, showing that the use of G <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">0</sup> <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">P</sub> model with quantile partition inicialization provide good segmentation results with few iterations.