基于块匹配学习字典的图像去噪算法  

Power Image Denoising Based on Block Matching and Learning Dictionary

作　　者：李松 陈勇 王云辉 谢珉 杨永昆 LI Song;CHEN Yong;WANG Yunhui;XIE Min;YANG Yongkun(Honghe Power Supply Bureau,Yunnan Power Grid Co.,Ltd.,Honghe Yunnan 661100,China;Yunnan Power Grid Co.,Ltd.,Kunming Yunnan 650011,China)

机构地区：[1]云南电网有限责任公司红河供电局,云南红河661100 [2]云南电网有限责任公司,云南昆明650011

出　　处：《信息与电脑》2023年第16期83-87,共5页Information & Computer

基　　金：云南电网有限责任公司科技项目(项目编号:YNKJXM20210149)。

摘　　要：文章提出基于块匹配学习字典的电力图像去噪算法。该算法将图像块间的相关性与稀疏表示相结合用于图像去噪。首先对图像块进行聚类,将相似的图像块分为一类,其次通过稀疏编码和字典学习建立每一类图像块的表示字典,以提高学习字典对于图像的表征能力,通过最小化块匹配去噪目标实现了图像去噪,最后对比各种去噪算法在不同噪音水平下对电力设备图像的去噪效果。实验结果表明,所提出算法在不同噪音水平下,均表现出较好的去噪性能,峰值信噪比(Peak Signal to Noise Ratio,PSNR)和结构相似度(Structural Similarity,SSIM)高于传统去噪算法。The article proposes a power image denoising algorithm based on block matching learning dictionary.This algorithm combines the correlation between image blocks with sparse representation for image denoising.Firstly,image blocks are clustered and similar image blocks are classified into one class.Secondly,a representation dictionary for each class of image blocks is established through sparse encoding and dictionary learning to improve the representation ability of the learning dictionary for images.Image denoising is achieved by minimizing block matching denoising targets.Finally,the denoising effects of various denoising algorithms on power equipment images at different noise levels are compared.The experimental results show that the proposed algorithm exhibits good denoising performance at different noise levels,with higher Peak Signal to Noise Ratio(PSNR)and Structural Similarity(SSIM)compared to traditional denoising algorithms.

关 键 词：块匹配 学习字典 电力图像 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]