DCVS系统中基于双稀疏冗余字典的高性能解码方法

为改善分布式压缩视频感知(distributed compressive video sensing,DCVS)系统的视频帧图像重构质量,以实时视频传输为应用场景,提出了一种基于双重稀疏模型的图像解码算法。解码端由相邻的已重构关键帧产生边信息(side information,SI);根据双重稀疏模型思想,分离样本图像小波域下不同尺度的子带,分别使用K均值奇异值分解(K-means singular value decomposition,K-SVD)算法得到具有多尺度特性的冗余字典,结合梯度投影稀疏重建(gradient pursuit for sparse reconstruction, GPSR)算法,完成对非关键帧的重构。仿真结果表明,在相同压缩率下,相比传统K-SVD字典训练方法, 本文所提出的方法对应的视频帧图像重构峰值信噪比(peak signal to noise ratio,PSNR)可获得0.5~1.5 dB以上的增益。

High-performance decoding method based on double sparse dictionary in DCVS system

To improve the quality of video frame image-reconstruction in the distributed compressive video sensing(DCVS) system, the continuous real-time video transmission is used as scenarios, an decoding algorithm based on the double sparse dictionary is proposed. Side information(SI) is generated by the adjacent key-frame. Separate sub-bands with different scales of the sample images under wavelet domain, then use the K-SVD algorithm to get different redundant dictionaries with multi-scale properties. Finally, these dictionaries and gradient pursuit for sparse reconstruction(GPSR) algorithm are combined to reconstruct the non-key frames. Simulation results show that at the same compression ratio, the peak signal to noise ratio (PSNR) of the proposed method can obtain more than 0.5~1.5 dB gain than traditional methods.