A Comparison of Statistical Segmentation Techniques for Polarimetric Sar: Region Growing versus Simulated Annealing

In this paper, two polarimetric segmentation techniques for polarimetric SAR images are compared. They are both based on the maximum generalised likelihood approach and on a Wishart distribution model. The first technique, named POLSEGANN, is based on a global likelihood approach and on the simulated annealing maximization technique, while the second one (POL MUM) is based on a Maximum Likelihood (ML) Split-Merge test between adjacent regions, and on a region growing scheme. Both techniques exploit the properties of the covariance matrix of the data, but they proceed with very different approaches to identify the widest possible homogeneous segments. The comparison of the two techniques is performed both on a wide set of simulated images and on real data. Results are evaluated qualitatively and quantitatively, considering the accuracy achieved in the classification of the segmented images and the statistical characteristics of the ratio image. We show that POLSEGANN provides more accurate classification results and a better identification of small regions, while POL MUM provides an accurate statistical reconstruction of the original image, and the identification of large homogeneous regions.