基于改进加权核范数最小化的图像去噪算法研究  

Image Denoising Based on Improved Weighted Nuclear Norm Minimization

作　　者：史凯特 孙浩东 董秀芬 马鹏阁[2] 漆召兵 张亚平 秦晓科 SHI Kaite;SUN Haodong;DONG Xiufen;MA Pengge;QI Zhaobing;ZHANG Yaping;QIN Xiaoke(China Three Gorges Corporation Research Institute,Beijing 101000,China;Zhengzhou University of Aeronautics Intelligent Engineering College,Zhengzhou 450000,China;China Three Gorges Corporation Strategy and Development Research Center,Beijing 100000,China;Three Gorges New Energy Offshore Wind Power Operation and Maintenance Jiangsu Co.Ltd,Yancheng 224000,China;Luoyang Electro-Optical Equipment Research Institute China Aviation Industry Corporation,Luoyang 471000,China)

机构地区：[1]中国长江三峡集团有限公司科学技术研究院,北京101000 [2]郑州航空工业管理学院智能工程学院,郑州450000 [3]中国长江三峡集团有限公司战略与发展研究中心,北京100000 [4]三峡新能源海上风电运维江苏有限公司,江苏盐城224000 [5]中国航空工业集团公司洛阳电光设备研究所,河南洛阳471000

出　　处：《电光与控制》2024年第7期48-52,共5页Electronics Optics & Control

基　　金：国家自然科学基金民航联合基金重点项目(U1833203);中国长江三峡集团有限公司科研项目(202003035)。

摘　　要：针对彩色图像的高斯噪声,在加权核范数最小化(WNNM)框架下,提出了一种对于彩色图像去噪的多信道(MC)优化模式。首先,选择多种类型噪声计算量,利用信道的冗余特性将RGB补丁连接起来,再引入权重矩阵以协调3种信道的图像保真性。把所提的MC-WNNM模型转换成线性等式约束现象,并采用交替位置乘子法(ADMM)解决。每个变量更新步骤都具有其封闭解,并能保证收敛性。基于真实用于无人机目标识别的彩色图像添加噪声进行仿真实验,实验结果说明,该算法相较现有的BM3D方法及WNNM方法具有明显优势。Aiming at removal of Gaussian noise from color images a Multi-Channel(MC)optimization model for color image denoising is proposed in the framework of Weighted Nuclear Norm Minimization(WNNM).First multiple types of noise computation are selected RGB patches are connected by using the redundancy property of the channels.Then a weight matrix is introduced to reconcile the image fidelity of the three channels.The given MC-WNNM model is converted into a linear equation constrained phenomenon and is solved by using the Alternating Direction Method of Multipliers(ADMM).Each variable update step has its own closed solution and convergence is guaranteed.Simulation experiments based on real color images for UAV target identification with added noise show that the method has significant advantages over the existing BM3D and WNNM methods.

关 键 词：图像去噪 高斯噪声 加权核范数最小化 交替方向乘子法 

分 类 号：TP751.1[自动化与计算机技术—检测技术与自动化装置] TN911.73[自动化与计算机技术—控制科学与工程]