基于半像素运动估计的DCVS残差重建算法

Residual reconstruction algorithm based on half pixel interpolation motion estimation for distributed compressive video sensing

针对分布式视频压缩感知中非关键帧重建质量较差的问题进行了研究,提出了一种基于半像素插值技术的残差多假设块迭代重建算法。该算法先对重建后的相邻关键帧进行半像素插值运动估计生成当前非关键帧的边信息;然后在测量域内对非关键帧与边信息求残差,并对残差进行多假设重建;最后将重建的残差与边信息进行融合得到重建的非关键帧。仿真结果表明,基于半像素插值的改进残差多假设块迭代重建算法比多假设迭代重建算法重建质量平均提高了0.3~1.4 dB,提高了非关键帧的重建质量。

This paper studied the problem of the poor quality of non-key frame reconstruction in distributed compressive video sensing and proposed a residual reconstruction algorithm based on half pixel interpolation for the problem. First, it obtained a side information of the current non-key frame by dealing with the reconstruction of two adjacent key frames based on half pixel interpolation. Second, it implemented a reconstruction of the difference between non-key frame and side information （SI） in measurement domain using multi hypothesis predictions. Third, it obtained the result of non-key frame by dealing with the SI and the reconstructed difference. Experimental results reveal that, compared to the residual reconstruction algorithm based on multi hypothesis predictions, the proposed residual reconstruction algorithm improves the quality of non-key frame reconstruc- tion by O. 3 dB to 1.4 dB on the average.