关键词: 图像去噪, 椒盐噪声, 非局部平均, 相似权, 迭代滤波

Abstract:

According to the fact that the nonlocal means (NLM) method cannot adequately remove noise from the images corrupted by salt-and-pepper noise, we extend the NLM to remove salt-and-pepper noise by introducing the noise detection results. At the noise detection stage, we divide pixels into two categories: noisy and noise-free pixels, depending on two extreme values Lmin and Lmax. At the filtering stage, noise-free pixels remain unchanged, while for each noise pixel, if the adaptive filtering window does not contain any noise-free pixel, we regard the current noise pixel located in image uniform regions composed of noise-free pixels with the same gray value Lmin or Lmax. And then the calculated statistics is used as the restored value. Otherwise, we employ the improved NLM filter for noise removal. The joint noise detection mask in the proposed method can avoid the influence of noise pixels on calculating similar weights in the presence of noise pixels, and only noise-free pixels are used for the weighted average. In addition, the iterative filtering scheme is used to remove noise of high-density. Experimental results demonstrate the effectiveness of the proposed filter  even though  its computational complexity is still high.

Key words: image denoising, salt-and-pepper noise, nonlocal means, similarity weight, iterative filtering