Assessment of statistical models for clutter and target in SAR images

Accurate knowledge of statistical properties of SAR data plays an essential role in SAR image processing and understanding. Several studies have been made for discovering the relationship between the physical features and statistical properties of SAR data, and some statistical models for modeling SAR data having been proposed and studied. In this paper, we focused on four often used statistical models: the Weibull, Log-normal, Gamma and K distributions. These models are used to fit target and clutter regions in SAR data provided by MSTAR, and through three different goodness-of-fit tests, we assess the performance of the four statistical models for modeling the clutter and target in the SAR images. The results show that K distribution performs best and Log-normal performs worst for modeling clutter region, on the other hand, Log-normal distribution performs best while K distribution performs worst for modeling target region.