基于运动对齐预测模型的分布式视频压缩感知重构  

Distributed Video Compressive Sensing Reconstruction Based on Motion-Aligned Predictive Model

作　　者：张健[1] 董育宁[1] 

机构地区：[1]南京邮电大学通信与信息工程学院,江苏南京210003

出　　处：《南京邮电大学学报（自然科学版）》2014年第4期63-71,共9页Journal of Nanjing University of Posts and Telecommunications：Natural Science Edition

基　　金：国家自然科学基金(60972038;61271233);教育部博士点基金(20103223110001);江苏省普通高校研究生科研创新计划(CXZZ12_0468)资助项目

摘　　要：为了提高分布式视频压缩感知(Distributed Video Compressive Sensing,DVCS)的率失真性能,文中提出根据视频非关键帧图像的时间相关性将帧内各块分为静止块与运动块两类,并对它们设定不同的测量率以提高压缩感知(Compressive Sensing,CS)捕获信息的效率。在重构过程中,提出运动对齐多假设预测模型进行重构,该预测模型在测量域内实现运动估计,并根据运动信息在参考帧内寻找到待重构块的若干候选匹配块,利用它们的线性加权和残差重构得到非关键帧图像的重构结果。仿真实验结果表明,文中所提出的DVCS重构算法能有效提升系统的率失真性能,与现有方法相比,在重构时间基本不变的情况下,获得更好的主客观视频重构质量。In order to improve the rate-distortion performance of distributed video compressive sensing （DVCS）,this paper proposes a new algorithm for classifying all blocks in a non-key video frame into static and motion blocks depending on their temporal correlations,and sets different measurement rates for improving the efficiency of information capture in compressive sensing（CS）.A motion-aligned multiple hypotheses predictive model is developed to reconstruct blocks in the process of reconstruction.This predictive model can realize the motion estimation in the measurement domain,use the motion information to find several matching blocks in the reference frame,and then obtain the reconstructed frame by using linear weighted sum of these candidates and residual reconstruction.Simulation results show that the proposed DVCS reconstruction algorithm can effectively improve the rate-distortion performance of the system and obtain better objective and subjective quality of the reconstructed video than existing approaches with the same reconstruction time.

关 键 词：分布式视频压缩感知 压缩感知 块分类 自适应测量率分配 运动对齐多假设预测模型 

分 类 号：TN911.73[电子电信—通信与信息系统]