分析了中值滤波及其改进型算法在处理高密度椒盐噪声时效果不理想的原因, 采用变分修复方法来去除高密度椒盐噪声, 基于现有的全变差修复模型提出了非局部全变差修复模型。 该模型利用椒盐噪声特点(均匀分布、灰度值为0或255), 将噪声点看成是图像中遗失或是破损的点, 首先在图像中寻找与噪声点邻域相似的区域, 将相似区域的中心像素作为噪声点新的邻域然后对其插值, 把图像降噪问题转化为图像修复问题, 从而达到去除高密度噪声的目的。实验结果表明: 该模型对噪声密度为90%的彩色和灰度图像去噪后, 其峰值信噪比为2285和2877, 在客观评价标准方面优于中值滤波及其改进型算法。该模型能有效去除高密度下的椒盐噪声并较好地恢复图像细节, 为图像去除高密度噪声提供了一种新的途径。

The reasons of ineffectiveness of median filtering and its improved algorithm for eliminating the high-density salt-and-pepper noise are analyzed. A variational inpainting method is adopted to remove the high-density salt-and-pepper noise, and a inpainting model of Non-local Total Variation(NL-TV) based on the existing model of Total Variation(TV) is proposed in this article. In the NL-TV model based on the characteristics of salt-and-pepper noise(uniform distribution and the gray value of 0 or 255), we view the noise points as the lost or damaged points of an image to find the districts similar to the neighborhoods of noise points in an image, and then interpolate the noise points by taking the central pixel in a similar district as a new neighborhood of noise points. By this method, we transform the problem of image denoising into a problem of image restoration to remove the high-density noise. The experimental results show that the Peak Signal to Noise Ratios(PSNRs) are 2285 and 2877 after removing the noise for a color and gray-scale image with 90% of noise density, which is better than the results obtained by median filter and its improved algorithm in terms of the objective evaluation criteria. Using this model, we can effectively remove the high-density salt-and-pepper noise and restore the image details better, which provides a new approach to remove the high-density noise.