基于非局部结构张量的图像去噪算法

Image denoising algorithm based on non-local structure tensor

针对非局部均值滤波在计量像素补丁之间的相似度不够准确的问题,该文先引入各向同性中心凹算子对像素补丁进行预处理,以抑制噪声的不良影响,随后再进行相似度计算,且为了充分利用图像中像素点的几何结构等细节信息,进一步提出了基于非局部结构张量的非局部均值去噪算法。实验结果表明,所提算法相较于传统非局部均值去噪(NLM)和基于结构张量的非局部均值去噪(STNL)在峰值信噪比上分别提高了6.57%、2.1%,同时视觉呈现上也有明显提升。

Aiming at the problem that the similarity between neighborhood blocks in nonlocal means filtering is not accurate,a preprocessing for neighborhood blocks using isotropic foveation operators is introduced,which can restrain the bad effects of noise.Then the similarity calculation is carried out,and in order to be able to make full use of the geometric structure of pixels in the image,a nonlocal mean denoising algorithm based on nonlocal structure tensor is proposed.Compared with the traditional NonLocal Means(NLM)image denoising algorithm and the NLM based on structure tensor(STNL),the improved peak signal to noise ratio of the algorithm is 6.57%and 2.1%,respectively.And there is also a significant improvement in visual presentation.