基于分布式压缩感知的H.264帧间编码

Inter-frame Coding in H.264/AVC Based on Distributed Compressive Sensing

对视频信号压缩,国际标准算法H.264/AVC的帧间编码技术表现出显著的压缩性能,然而由于编码端依靠相邻帧间相关性进行压缩,造成其在丢包环境中较低的错误恢复能力。分布式压缩感知采用"独立编码-联合解码"框架,克服了编码端对帧间相关性的依赖,同时由于它对图像进行随机观测,对随机丢失具有较好的容错性。因此,本文将分布式压缩感知技术融合到H.264中,提出一种鲁棒的分布式压缩感知帧间编码方法:编码端仍采用H.264的运动估计并将运动矢量传送到解码端,同时用随机观测值代替运动估计残差并发送到解码端,解码端利用运动补偿的边信息进行压缩感知优化重构。为提高性能,同时提出对图像块的去均值观测、对观测值的合理打包以及解码端残差重构等方法。实验结果表明,在丢包环境下,本文提出的帧间编码模式比H.264具有更高的错误恢复能力。

The inter-frame coding in H. 264/AVC has excellent performance in coding efficiency, however,it is vulnerable to channels errors due to the exploitation of dependences between the adjacent frames at the encoder side. Based on the paradigm of separate encoding and joint decoding, distributed compressive sensing （DCS） offers a robust solution to the channel errors and at the same time, because of the random measurement, it has better error-resilience. In this paper, by integrating the respective characteristics of DCS and H. 264/AVC, inter-frame coding with DCS was proposed. The encoder replaces the residual with random measurements but still keeps the motion estimation and sends motion vector （MV） to the decoder; at the decoder side, the received measurements and side information （from motion-compensation） are both employed in joint CS decoding. In order to improve the performance, we also proposed some modifications such as the measurement algorithm removing the mean of image, the reasonable packaging method, the residual reconstruction and so on. The experimental results show that the proposed inter-frame coding is more robust against packet losses than that of H. 264/AVC in the environment of packet loss channels.