非局部低秩正则化视频压缩感知重构

Compressive video sensing reconstruction via nonlocal low-rank regularization

视频压缩感知在采样资源受限的视频采集领域具有重要研究意义,重构算法是视频压缩感知系统的关键技术.为了更好地从压缩采样数据中重构视频信号,提出一种基于全变分与非局部低秩正则化的视频重构算法,为视频重构提供一种新的思路.算法第1步考虑视频帧内和帧间的局部相关性,应用全变分模型作为先验约束得到初步恢复的视频帧;第2步考虑视频帧内与帧间的非局部自相似性,应用改进的非局部低秩正则化算法对其进一步重构,该步骤针对初步恢复的图像帧分块在本帧和关键帧中寻找相似块,构建低秩矩阵进行低秩正则化重构.仿真结果表明,所提出算法能够精确重构视频信号,相比主流的视频压缩感知重构算法具有更高的重构质量.

Compressive video sensing(CVS)has great research significance in the video acquisition system with limited sampling resources.This paper proposes a reconstruction algorithm based on total variation(TV)and nonlocal low-rank regularization(NLR-CS)to better reconstruct video signal from compressive sampled data.For this algorithm,the first step considers the local correlation within and between video frames,and applies TV as the prior constraint to obtain the initial recovered frame.In the second step,the improved NLR-CS algorithm is utilized to further reconstruct video frame considering the nonlocal self-similarity(NLSS).This step first blocks the initial recovered frame,finds similar blocks in the current frame and the key frames to construct low-rank matrix,then a low-ranking regularization reconstruction is performed.Experimental results show that the proposed algorithm can reconstruct video signals well,obtains higher video reconstruction accuracy than other CVS reconstruction algorithms.