摘要
为了解决基于低秩正则化的图像压缩感知重构算法不能充分利用图像局部梯度稀疏特性的问题,提出了一种基于低秩与全变差正则化的图像压缩感知重构算法.首先,通过图像块匹配法寻找结构相似的图像块,组成非局部相似块组;其次,联合相似块矩阵低秩与图像梯度稀疏先验组成正则化项,结合传统的压缩感知模型形成新模型;最后,采用交替方向乘子法实现图像的重构.测试图像为自然灰度图像,为了验证算法的有效性,从主观视觉和峰值信噪比两方面进行对比.试验结果表明,和基于低秩正则化的图像压缩感知算法相比,该算法在准确描述图像非局部自相似性结构特征的前提下提高了重构质量,重构的图像在峰值信噪比上平均提升1 d B.
To solve the problem that image compressive sensing reconstruction algorithm via nonlocal lowrank regularization could not adequately exploit the local gradient sparsity, an improved image construction algorithm was proposed based on low-rank and total variation regularization. The similar patches were found by image block matching method and formed into nonlocal similar patch groups. The regularization term of combining low-rank prior of nonlocal similarity patch groups with gradient was embedded into reconstruction model, which was solved by alternating direction multiplier method（ ADMM） to obtain the reconstructed image. The test images were gray scale images. To verify the proposed algorithm,the experimental results were compared by subjective vision and peak signal-to-noise ratio（ PSNR）. The experimental results show that compared with the algorithm via nonlocal low-rank,the proposed method can significantly improve the quality of reconstructed image with nonlocal self-similar structure precisely described,and the PSNR of reconstructed images is increased about 1 d B in average.
作者
杨桄
封磊
孙怀江
孙权森
YANG Guang FENG Lei SUN Huaifiang SUN Quansen(School of Computer Science and Engineering, Nanjing University of Science and Technology, Nanjing, .liangsu 210094, China)
出处
《江苏大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2017年第5期571-575,614,共6页
Journal of Jiangsu University：Natural Science Edition
基金
国家自然科学基金资助项目(61273251)
关键词
压缩感知
图像重构
全变差
低秩近似
交替方向乘子法
compressive sensing
image reconstruction
total variation
low-rank approximation
alternative direction multiplier method
