Grey relational analysis based adaptive smoothing parameter for non-local means image denoising

Abstract

In non-local means (NLM) algorithm used for suppression of noise in digital images, the choice of smoothing or decay parameter is a critical issue, which affects the performance of NLM algorithm by influencing the amount of smoothing. Generally, the smoothing parameter in NLM algorithm is kept fixed for all pixels in an image, which provides blurring effects near important image details such as edges, textures etc. in an image. This paper presents a grey relational analysis based adaptive non-local means (GRANLM) algorithm to select an adaptive smoothing parameter for each pixel. It considers the grade of relations between reference patch and adjacent patches in a defined region centred at pixel using grey relational analysis (GRA). Experimental results on various standard images for different noise levels show that the proposed algorithm outperforms the traditional NLM algorithm and other NLM variants in terms of peak signal to noise ratio (PSNR), structural similarity index measure (SSIM), visual quality and method noise.