保持纹理细节的自适应非局部均值图像降噪

Adaptive Nonlocal Means Image Denoising for Better Preservation of Textures

主邻域字典(principal neighborhood dictionaries,PND)非局部均值(nonlocal means,NLM)是一种基于主成分分析(principal component analysis,PCA)的有效图像降噪方法,但因其未能充分利用图像的内容结构信息,对纹理细节较多区域的降噪效果较差.改进PND方法,实现基于PCA的自适应非局部均值降噪.根据图像局部内容调整滤波参数h,得到动态变化的像素间相似权值.实验结果表明,该方法能更好地保留图像纹理和边缘信息,降噪效果优于非自适应的PND方法.

The principal neighborhood dictionaries （PND） nonlocal means （NLM） is an effective method for image denoising, which is based on principal component analysis （PCA）. However, it does not make full use of the image contents and is less effective in texture regions. This paper proposes modification to the PND method by adjusting a filter parameter h to achieve better accuracy. Experimental results show that the proposed outperforms PND method and can preserve more edge and texture details while achieving satisfactory denoising results.