Comments on randomly sampled non local means image filter

Abstract

In this work we comment the results presented in [1] regarding a random sampling approach of the Non Local Means (NLM) image denoising filter with respect to computational cost and denoising performance. We will show that although the approach is novel and mathematically revealing, the computation cost of the approach is higher, and the PSNR lower, compared to the classical version. Furthermore, we will present a probabilistic model to evaluate the performance of different versions of NLM and tune its parameters.