Denoising of single-look SAR images based on variance stabilization and nonlocal filters

Synthetic-aperture radar (SAR) imaging has become an efficient tool for obtaining and retrieving useful information about surfaces of Earth and other planets. However, the formed images suffer from speckle noise, especially if single-look observation mode is used. Then, filtering is often applied to improve image quality and provide better estimation of radar cross-section and other parameters of sensed scenes. Recently, a novel class of image filters has proved to be very successful in the removal of additive white Gaussian noise from natural images; these filters are based on nonlocal image modeling, i.e. they exploit the mutual self-similarity of image patches at different locations in the image. These filters have been shown in several benchmarks to significantly outperform all previous techniques. In this paper, we evaluate the performance of nonlocal filters applied to the denoising of single-look SAR images corrupted by speckle with a Rayleigh distribution, taking advantage of exact forward and inverse variance-stabilizing transformations. Numerical simulations demonstrate the success of this approach against several known despeckling methods.