多通道低秩先验的高斯模糊遥感图像复原算法  

Gaussian Blur Remote Sensing Image Restoration Algorithm Based on Multi-channel Low-rank Prior

作　　者：李喆 郭鑫 成丽波[1] 贾小宁 LI Zhe;GUO Xin;CHENG Li-bo;JIA Xiao-ning(School of Mathematics and Statistics,Changchun University of Science and Technology,Changchun Jilin 130022,China)

机构地区：[1]长春理工大学数学与统计学院,吉林长春130022

出　　处：《计算机仿真》2025年第2期204-209,共6页Computer Simulation

基　　金：国家自然科学基金(NO.12171054),吉林省教育厅科学技术研究项目(JJKH20230788KJ)。

摘　　要：针对高斯模糊导致的遥感图像模糊问题,在混合即插即用(hybridplug-and-play,H-PnP)算法的基础上,设计了多通道低秩先验的高斯模糊遥感图像复原算法。考虑到不同颜色通道中不同的噪声强度,引入权重矩阵来平衡不同颜色通道中的噪声,并结合多通道非局部相似块低秩先验和深度先验,对高斯模糊噪声遥感图像进行复原,最终利用交替迭代法对模型进行求解,复原出清晰图像。实验结果表明,上述算法对于叠加模糊和噪声的退化图像具有良好效果。选取CSR算法、INSR算法和H-PnP算法进行对比实验,在主观视觉效果、峰值信噪比和结构相似性方面均优于对比算法。Aiming at the blur problem of remote sensing images caused by Gaussian blur,this paper designs a multi-channel low-rank prior Gaussian blur remote sensing image restoration algorithm based on a hybrid plug-andplay(H-PnP)algorithm.Considering the different noise intensities in different color channels,the weight matrix is introduced to balance the noise in different color channels,and the Gaussian blur noise remote sensing image is restored by combining the multi-channel non-local similar block low-rank prior and deep prior.Finally,the alternate iteration method is used to solve the model and the clear image is restored.Experimental results show that this algorithm has a good effect on degraded images with superimposed blur and noise.This paper selects the GSR algorithm,INSR algorithm and H-PnP algorithm to be comparative experiments.The results illustrate that the algorithm is superior to comparative algorithms in aspects of subjective visual effect,peak signal-to-noise ratio and structural similarity.

关 键 词：深度先验 多通道低秩先验 交替方向乘子法 图像去模糊 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]