改进非局部均值各向异性扩散图像去噪算法  

Improved Image Denoising Algorithm Based on Non-Local Means Anisotropic Diffusion

作　　者：王磊[1] 王敏 张鹏程[1] 任时磊 高晓玲[2] 桂志国[1] WANG Lei;WANG Min;ZHANG Pengcheng;REN Shilei;GAO Xiaoling;GUI Zhiguo(Shanxi Provincial Key Laboratory for Biomedical Imaging and Big Data, North University of China, Taiyuan 030051, China;Linfen People's Hospital, Linfen 041000, China)

机构地区：[1]中北大学生物医学成像与影像大数据山西省重点实验室,山西太原030051 [2]临汾市人民医院,山西临汾041000

出　　处：《测试技术学报》2021年第5期436-442,449,共8页Journal of Test and Measurement Technology

基　　金：国家自然科学基金资助项目(61671413,61801438);中北大学青年学术带头人项目(QX201801);山西省高校科技创新项目(2020L0282);临汾市科技计划项目(1911)。

摘　　要：针对各向异性扩散算法的不稳定性在去噪过程中易产生阶梯效应,导致纹理细节不能较好保留的缺点,提出一种改进的非局部均值各向异性扩散图像去噪算法.所提算法将非局部均值引入迭代过程,利用图像中的大量相似冗余信息,以中心像素邻域图像块作为像素点的相似性度量,对图像进行合理的描述;同时在扩散系数函数中,将图像梯度结合残差图像中局部能量算子,自适应选择扩散系数.结果表明,所提算法在有效平滑噪声的情况下,减弱了阶梯效应,保留了更多的纹理细节,与各向异性扩散算法、自适应选择扩散去噪算法和基于非局部均值理论的各向异性扩散算法相比,峰值信噪比和结构相似度有较好提升.Due to the instability of anisotropic diffusion algorithm(P-M model),it is easy to produce staircase effect in the process of denoising,which leads to the fact that texture details can not be well preserved,An improved P-M model based on Non-Local Means(NLM)image denoising algorithm is proposed.In the proposed algorithm,NLM is introduced into the iterative process,and a large number of similar redundant information in the image is used to describe the image more reasonably by taking the image block in the neighborhood of the center pixel as the similarity measure of the pixel.At the same time,in the diffusion coefficient function,the image gradient is combined with the local energy operator in the residual image to select the diffusion coefficient adaptively.The experimental results show that the proposed algorithm can effectively smooth the noise,weaken the staircase effect and retain more texture details.Compared with the classical P-M model algorithm,adaptive selection diffusion denoising algorithm and P-M model based on NLM theory,the peak signal-to-noise ratio and structural similarity are also improved.

关 键 词：各向异性扩散 非局部均值 纹理检测 残差图像 图像去噪 

分 类 号：TP391.7[自动化与计算机技术—计算机应用技术]