On Hierarchical Segmentation of High Resolution PolSAR Data

Abstract:
Segmentation of SAR (Synthetic Aperture Radar) and PolSAR (Polarimetric SAR) images is a challenging task, both because such images are strongly textured and because of the relatively complicated stochastic model that is assumed. Such stochastic models may vary from classical Gaussian to the more advanced SIRVs. Since contour approaches are higly unreliable for textured images, multi-resolution techniques, among which hierarchical segmentation accounts, gained importance in PolSAR segmentation. However, finding the optimal number of segments, when the hierarchical algorithm should be stopped, remain a subjective task. This paper reports some results concerning this latter question.