Noise Suppression by Discontinuity Indicator Controlled Non-local Means Method

Abstract

The non-local means (NLM) method and its variances have been proved to be effective for noise suppression. However, the traditional NLM method and its variances cannot guarantee to obtain a global optimum solution owing to a globally fixed bandwidth parameter for the entire image is used when computing similarity weight function. To address this problem, this paper proposes an adaptive NLM method based on a novel discontinuity indicator, which can get a good tradeoff between edge preservation and noise reduction. In our method, a novel discontinuity indicator based on the structure tensor is proposed, which can effectively distinguish edges from noises and smooth regions. Furthermore, the bandwidth parameter is adaptively chosen according to the proposed discontinuity indicator. As a result, the bandwidth parameter depends continuously on the local characteristic of each pixel. Experimental results demonstrate that our proposed method outperforms several mainstream methods.