A steering kernel based nonlocal-means method for image denoising

Abstract

The nonlocal-means (NLM) is a powerful method for image denoising which takes advantage of the redundancy of similar patches in the image. The steering kernel regression is a non-parametric estimation for image restoration that develops a data-adapted steering kernel based on local orientation estimate. In this paper, a steering kernel based nonlocal-means filter (SK-NLM) has been developed which not only exploits the self-similarity of the image, but also considering the structural information by the steering kernel. Experimental results show that the proposed method effectively improve the PSNR while preserving local structures.