基于残差重构的分布式视频压缩感知

Residual Reconstruction Based Distributed Compressed Video Sensing

为了改进分布式视频压缩感知方案的性能,提出了一种基于残差重构的分布式视频压缩感知方案。该方案在编码端逐帧独立进行测量,在解码端依靠视频信号的时域相关性提升重构信号质量。首先,对关键帧独立进行重构;其次,利用已重构关键帧做运动估计/运动补偿以生成非关键帧的边信息;接下来,对边信息采用与编码端相同的测量矩阵进行测量并计算测量残差值;最后,采用全变分最小化重构残差信号值并将其与边信息相加生成最终的重构图像。实验结果表明,在相同采样率下,与已有的分布式视频压缩感知方案相比,提出的方案可获得2.8 dB以上的峰值信噪比增益。

To improve the performance of Distributed Compressed Video Sensing（ DCVS）, a residual reconstruc- tion based DCVS framework is proposed. The proposed framework samples each video frame independently at the encoder. However, it recovers frames at the decoder by exploiting inter frame correlation. Firstly, the key frame of a Group of Pictures（GOP） is independently reconstructed. Secondly, Side Information（SI） is generated by performing bi-directional Motion Estimation（ME） and Motion Compensation（MC） through the reconstructed key frames. Afterwards, the generated SI frame is sampled by the same matrix as the one at the encoder, and the measurement of SI is used to calculate the residual of measurement. Finally, total variation minimization is ap- plied to reconstruct the residual signal, and the output frame is formed by adding SI to the residual signal. Ex- perimental results show that compared with the existing DCVS method, the proposed one can get more than 2.8 dB Peak Signal to Noise Ratio（PSNR） increment at the same sampling rate.