联合时空特征的视频分块压缩感知重构

为了提高视频压缩感知(CS)重构算法的率失真性能,该文提出利用视频的时空特征进行联合重构。为了不引入过多的复杂度,采集端以固定采样率对帧内各块进行测量;重构端则在最小全变差(TV)重构模型的基础上,分别加入利用时空自回归(AR)模型和多假设(MH)模型所形成的正则化项,以提高预测-残差重构的性能。另外,考虑到视频源的统计特性在时空域中是动态变化的,讨论了5种不同的帧间预测模式对重构精度和重构计算复杂度的影响。仿真实验表明,所提出的重构算法能够以一定的计算复杂度为代价有效地改善视频重构质量,且在关键帧采样率高于非关键帧的情形下,帧间预测模式的改善也可一定程度上提高视频重构质量。

Block Compressed Sensing Reconstruction of Video Combined with Temporal-spatial Characteristics

To improve the rate-distortion performance of video Compressed Sensing (CS) reconstruction, the temporal-spatial characteristics of video are used to jointly recover the video signal in this paper. At the collection terminal, each block in a single-frame is measured at the fixed sampling rates to advoid excessive complexity. At the reconstruction terminal, two regularization terms are respectively added to the minimum Total Variation (TV) reconstruction model to advance the performance of prediction-residual reconstruction, and the terms are constructed in terms of temporal-spatial Auto-Regressive (AR) model and Multiple Hypothesis (MH) model. In addition, considering that the statistics of video source are dynamically varying in spatial and temporal domain, it is discussed how the five different inter-prediction modes impact on precision and computational complexity of reconstruction. Simulation results show that the proposed algorithms effectively improve the quality of reconstructed video at the cost of the computational complexity , and the improvement of inter-prediction mode enhances reconstruction quality in some extent.