Automatic Parameter Selection for the PolSAR Bilateral Filter

Abstract:
In this paper we propose an automatic parameter selection method for the polarimetric bilateral filter. Our approach is based on the theory of unbiased risk estimation and allows a per-pixel selection of the spatial and radiometric scales of the filter. The algorithm is tested on experimentally acquired data and is compared to filtering results where one or both parameters are fixed for the whole image. It is shown that our method improves the spatial adaptivity of the filter while avoiding a manual tuning of the parameters.