Non-Local Means Filters for Full Polarimetric Synthetic Aperture Radar Images with Stochastic Distances

Abstract

In this paper, we present a Non-Local Means filtering method for PolSAR (Polarimetric Synthetic Aperture Radar) imagery. We adopted three stochastic distances to measure the similarity of samples. We obtain p-values using the asymptotic distribution of test statistics based on these distances. Then we use this similarity evidence as input for a smooth activating function that yields the weights of local convolution matrices. Non-local filters are computationally demanding, so we provide an efficient implementation that allows us to experiment with different settings and find optimal parameters.