一种最佳线性估计与多假设结合的分布式视频压缩感知重构算法

Distributed Video Compressed Sensing Reconstruction Algorithm Combining Optimal Linear Estimation with Multiple Hypotheses

摘要: 针对分布式视频压缩感知系统存在编码端的低采样率易导致解码端重构效果不理想的问题，综合考虑视频帧间的时空相关性以及帧内不同图像块所具有的不同块特征，提出了一种最佳线性估计与多假设预测相结合的分布式视频压缩感知重构算法。算法在编码端采用分块压缩感知进行随机测量，在解码端增加了相似判别、测量值补充以及平滑判别三种机制。通过此三种机制对非关键帧块进行细化分类，并根据分类结果对不同的图像块采用不同的测量值补充及重构策略，进而提高重构质量。在标准视频序列集上的仿真实验结果表明，本文提出的算法重构视频信号的峰值信噪比(PSNR)比传统的多假设预测重构算法平均高出2~3 dB，尤其在编码端测量率低于0.2的情况下，PSNR高出4~5 dB。

Abstract: In the traditional distributed video compressed sensing system, low sampling rate in the encoding side often leads to the problem that the reconstruction in the decoding side is unsatisfactory. In order to solve this shortcoming, a distributed video compressed sensing reconstruction algorithm combining the optimal linear estimation with the multiple hypotheses prediction is proposed, which takes into account the spatial-temporal correlation between video frames and the different block features of different image blocks in the video frame. The block-based compressed sensing method is used to measure the video frames at the encoding side. At the decoding side, three kinds of mechanisms are added, such as similarity discrimination, measurement value supplement and smooth discrimination, which can be used to classify the blocks of non-critical frame. Then according to the classification results, the different blocks of non-critical frame use different measurement supplement strategies and can be reconstructed by different reconstruction strategies with an aim to improve the reconstruction quality. Experimental results on the public commonly used video test sequences demonstrate that the proposed algorithm outperforms the traditional multiple hypotheses predictive reconstruction algorithm by 2~3 dB (PSNR), by 4~5 dB in the case of the sampling rate below 0.2 especially.