基于S_(1/2)范数的非局部视频去噪  

Nonlocal video denoising based on S_(1/2) matrix norm

作　　者：张金利[1,2] 李敏[1] 何玉杰[1] 

机构地区：[1]第二炮兵工程大学初级指挥学院,陕西西安710025 [2]武警工程大学信息工程系,陕西西安710086

出　　处：《光电子．激光》2015年第5期951-959,共9页Journal of Optoelectronics·Laser

基　　金：国家自然科学基金项目(61102170)资助项目

摘　　要：为了去除视频中的高斯噪声及脉冲噪声,提出了一种基于S1/2矩阵范数的非局部视频去噪算法。首先,在视频数据中利用非局部块匹配的钻石搜索算法搜寻与参考图像块最相似的图像块组;然后,将搜寻到的相似图像块组列向量化后组合成的矩阵进行基于S1/2范数的低秩和稀疏分解,分解后的低秩成分视为原视频场景信息,稀疏成分视为视频中存在的随机值脉冲噪声及异常值数据;最后,由低秩矩阵恢复的各图像块数据经过加权平均后作为参考图像块的去噪估计值,进而求得视频各帧图像的去噪估计值。实验结果表明,本文方法能够有效去除视频中的高斯噪声和脉冲噪声,相比同类算法,去噪后的视频无论在视觉质量上还是客观评价指标上都有明显的优势。In order to remove Gaussian noise and impulse noise from video data, a nonlocal video denoising algorithm based on S1/2 matrix norm is proposed. Firstly, using a diamond search algorithm for fast patch-matching,some patches similar to the given reference patch are found and collected in the video data. Secondly,all of the columns of similarity patches are recombined to form a new matrix and the new matrix is decomposed into a low rank matrix and a sparse matrix based on $1/2 matrix norm. The low rank matrix represents the scene information data of the original video and the sparse matrix represents the impulse noise data and outliers existing in the noisy video. Lastly, the estimated values of a denoised reference patch are determined by taking the weighted average of all the data recovered from the low rank matrix,and the estimated values of the denoised frame are got b^ased on the combinations of all the recovered reference patches in a frame. Experimental results show that the proposed scheme can effec- tively remove Gaussian noise and impulse noise from the video. Compared with two existing state-of-art algorithms, the proposed algorithm has noticeable superiority in both visual effect and objective evalua- tion.

关 键 词：视频去噪 S1/2范数 低秩矩阵 稀疏矩阵 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]