Parameter Estimation of Generalized Gamma Distribution Toward SAR Image Processing

Statistical modeling of synthetic aperture radar (SAR) data is a crucial step in SAR image processing. In this context, the generalized gamma (GGamma) distribution, which generalizes many common distributions, has been applied to model SAR image statistics. Parameter estimation remains, however, a challenging step that conditions the quality of model fitting to data and, thus, of the required processing. In this article, we propose a novel parameter estimation method for GGamma distribution in the log-domain, named as the maximum likelihood and logarithmic cumulants (ML–LC) method. The ML–LC method constructs a novel scale-independent shape parameter estimator in the log-domain based on the Mellin transform and maximum-likelihood estimation and estimates the distribution parameters based on the multistart local search, gradient descent, and bisection methods, rather than solving the system of highly nonlinear equations in the traditional estimations. The ML–LC method is able to estimate the GGamma distribution parameters more accurately. To assess the performance of our estimation method, we perform the goodness-of-fit test on simulated data and real SAR images. In addition, we apply the ML–LC method in some SAR image processing tasks covering image segmentation, classification, and change detection. The results obtained confirm the interest of the proposed ML–LC method.