Active matching for patch adaptivity in nonlocal means image denoising

Abstract

Baseline nonlocal means denoising scheme may be improved by incorporating more adaptivity, like locally varying filtering window, smoothing constants and patch size or shape. In this paper, we presents a novel adaptive nonlocal means filtering scheme, the key idea of which is that before computing the similarity between two pixels, active matching is performed to determine optimally matched patch shape and size. Systematic analysis and detailed simulation results show that the proposed algorithm achieves excellent trade-off between bias and variance, and obtains superior denoising performance compared with the state of the art.