一种改进的非局部均值去噪算法  

An Improved Non-local Means Denoising Algorithm

作　　者：何春 宋国琴 郭科[2] HE Chun;SONG Guoqin;GUO Ke(Education and Information Technology Center,China West Normal University,Nanchong Sichuan 637009,China;Geomathematics Key Laboratory of Sichuan Province,Chengdu University of Technology,Chengdu Sichuan 610059,China)

机构地区：[1]西华师范大学教育信息技术中心,四川南充637009 [2]成都理工大学数学地质四川省重点实验室,成都610059

出　　处：《西华师范大学学报（自然科学版）》2020年第4期421-428,共8页Journal of China West Normal University(Natural Sciences)

基　　金：数学地质四川省重点实验室开放基金项目(scsxdz2018yb08);西华师范大学基本科研业务费项目(19D043);西华师范大学英才科研基金项目(17YC185)。

摘　　要：针对传统非局部均值算法耗时长,权重分配不够合理以及利用欧氏距离计算邻域之间相似性时未能突出中心像素点作用的问题,提出一种改进的非局部均值去噪算法。该算法利用积分图像原理,将对图像中每一个像素进行邻域相似性处理,改进为对图像矩阵进行统一处理,并且利用不同半径的高斯核函数累加从而加强了邻域中心点的权重地位,最后通过高斯函数幂运算对度量邻域之间相似度的权重函数进行了改进。实验表明,本文提出的算法大大缩短了算法运行时间,并且能更准确地度量邻域相似度,能更好地保留图像的边缘和细节信息,有效地提升了图像去噪效果。In this paper,an improved non-local mean denoising algorithm is proposed to solve the shortcomings of traditional non-local mean algorithm,including long time consumption,unreasonable weight allocation and failure to highlight the role of the central pixel when calculating the similarity between neighborhoods by Euclidean distance.The algorithm employs the principle of integral image to process the image matrix uniformly rather than the neighborhood similarity of each pixel,and the Gaussian kernel function of different radii is added up to strengthen the weight position of the center point in the neighborhood.Finally,the weight function that measures the similarity between neighborhoods is improved by Gaussian function power operation.Experiments show that the proposed algorithm has effectively improved the image denoising effect by shortening greatly the running time of the algorithm,measuring the neighborhood similarity more accurately,and retaining the edge and detail information of the image better.

关 键 词：非局部均值 高斯核函数 权重函数 图像邻域 图像去噪 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]