中文摘要：

现有的分块视频压缩感知在获取边信息时,通常对所有图像块均采用固定权值边信息合成方法,该方法忽略了不同图像块之间相关度不同的问题。针对这一问题,根据贝叶斯压缩感知和运动估计理论,提出了一种基于块的分类加权边信息生成方案的分布式视频解码方法。在解码端利用相邻关键帧中不同块的相关度差异,对相邻关键帧进行基于块的分类加权运动估计,生成边信息,进而完成非关键帧的重构。考虑到加权系数的大小取决于相邻关键帧对应块的相关度,所采用的重建算法是基于TSW-CS模型的贝叶斯压缩感知重构算法。分别采用固定权值边信息生成方法和分类加权边信息生成方法对不同视频序列进行了实验对比,实验结果表明,采用分类加权边信息方法生成的视频重建PSNR值比固定权值边信息生成方法平均提高了0.2~0.5 d B,所采用的解码方法可有效地提高视频压缩感知重构质量。

英文摘要：

For most of those existing block-based compressed sensing of video, the fixed weight side information generation method is usually utilized for all blocks, which underestimates the problem of the difference of correlation between different blocks. To address this issue, a classified weighted side information generation method with block for distributed video decoding has been proposed according to the Bayesian compressive sensing and motion estimation theory. In the decoding side, the different correlations of neighboring key-frames has been used to generate side information by taking classified weighted motion estimation with block to different block of key-frame, then the reconstruction of the non-key-frame is completed. Considering that weighting coefficient depends on the size of the adjacent frames relevance, the Bayesian compressive sensing reconstruction algorithm is adopted based on TSW-CS model. Fixed weight side information generation method and the proposed method are used in experiments for comparison with various video sequences. The experimental results show that the PSNR of reconstructed video of proposed side information generation method has been averagely improved 0. 2-0.5 dB, higher than fixed weight method. The restructure quality of video compression sensing has been effectively improved by pro- posed algorithm.