Bounded Non-Local Means for Fast and Effective Image Denoising

摘要

Non-Local Means (NLM) is a powerful but computationally expensive image denoising algorithm, which estimates a noiseless pixel as a weighted average across a large surrounding region whereby pixels centered at more similar patches are given higher weights. In this paper, we propose a method aimed at improving the computational efficiency of NLM by quick pre-selection of dissimilar patches thanks to a rapidly computable upper bound of the weighting function. Unlike previous approaches, our technique mathematically guarantees all highly correlated patches to be accounted for while discarding dissimilar ones, this providing not only faster speed but improved denoising too.