残差分布式视频压缩感知*

压缩感知(CS)是一种能同时进行数据采集和压缩的新理论,为简化编码算法提供了依据,同时,分布式视频编码(DVC)为低复杂度的视频编码提供了思路。因此,通过整合DVC和CS各自的特性以构建编码简单的视频编码框架,并采用残差技术来提高系统性能,最终提出了一种残差分布式视频压缩感知(RDCVS)算法:对关键帧进行传统的帧内编、解码;而对非关键帧,编码端采用一种基于残差联合稀疏模型的随机观测,解码端利用边信息和改进的梯度投影重建(GPSR)算法进行优化重构。由于将运动估计和变换编码等复杂度较高的运算转移到解码端进行,因而RDCVS保持了低复杂度的编码特性。实验结果表明,RDCVS算法比参考方案的恢复质量提高了2～3 dB。

Residual distributed compressive video sensing

Compressive sensing (CS) is a new theory which states that the compression can be achieved in the process of data acquisition, gives strong supports for simplifying the acquisition algorithms. Meanwhile, DVC provides a basis for low-complexity video coding. Hence,this paper addressed the problem of low-complexity video coding by integrating the respective characteristic of DVC and CS.It proposed a new residual distributed compressive video sensing (RDCVS) using the residual coding technology. At the encoder,it just encoded each key frame by traditional intra-frame model and measured the other non-key frames randomly based on a residual joint sparse model. At the decoder,it implemented the efficient reconstruction using side information and the modified GPSR (gradient projection for sparse reconstruction). Some computation-consuming algorithms, such as motion estimation and transform coding, are moved to the decoder, the low-complexity encoding is therefore preserved in RDCVS. The experimental results show that RDCVS achieves 23 dB improvements in PSNR compared to the referenced scheme.