一种复合局部与非局部梯度的图像去噪变分模型  

A Total Variation Model for Image Denoising Based on Local and Nonlocal Gradient

作　　者：吴洋 杨平先[1] 黄坤超 陈明举[1] 张雷 WU Yang;YANG Pingxian;HUANG Kunchao;CHEN Mingju;ZHANG Lei(College of Information Engineering,Sichuan University of Science and Engineering,Zigong 643000,China;Southwest China Institute of Electronic Technology,Chengdu 610036,China;AVIC Chengdu Aircraft Industrial(Group) Co. ,Ltd. ,Chengdu 610092,China)

机构地区：[1]四川理工学院自动化与电子信息学院,四川自贡643000 [2]中国西南电子技术研究所,成都6100362 [3]航空工业成都飞机工业(集团)有限责任公司,成都610092

出　　处：《电讯技术》2018年第12期1452-1457,共6页Telecommunication Engineering

基　　金：四川省教育厅项目(17ZB0302);人工智能四川省重点实验室开放基金项目(2016RYY02)

摘　　要：局部变分有效地增强图像的轮廓信息,但不可避免地模糊图像的细节并在平滑区域产生阶梯效应。非局部变分能有效重构图像的纹理信息,但同时会破坏图像的结构轮廓信息。考虑到局部与非局部变分的互补性,提出了一种基于图像局部梯度与非局部梯度的复合变分模型,并通过Bregman交替迭代极小化图像的局部梯度与非局部梯度的L1范数,使去噪后的图像在去除噪声的同时更好地保留图像的结构与细节信息。对比实验证明,提出的复合变分模型有效地利用了图像的局部变分与非局部变分的优点,在图像评价的主客观方面都表现出了更好的性能。Total variation(TV)is effective in sharping image structures while it tends to over smooth the image details and the staircasing effect inevitably occurs in image's smooth areas.Nonlocal total variation(NLTV)can keep fine details,but usually blurs image's structural edges.In consideration of the comple-mentarity of TV and NLTV,a new total variation for image denoising based on local and nonlocal gradient is proposed.The Bregman iteration is employed to minimize L1functional of local gradient and nonlocal gradient recursively.Therefore,the new total variation is efficient for image denoising while effectively preserving image textures.The comparison experimental results clearly demonstrate that the proposed total variation has the advantages of local total variation and nonlocal total variation,and also has better performance than some other state-of-the-art methods with regard to evaluation indices and visual quality.

关 键 词：图像去噪 变分模型 图像梯度 Bregman 迭代 

分 类 号：TN911.73[电子电信—通信与信息系统]