有损视频传输中的方向性结构与纹理重建

Directional structure and texture restoration in lossy video transmission

当被基于宏块预测的压缩方法压缩的视频码流通过有损信道传输的时候,丢包现象会导致错误沿着时域预测而传递。文中提出了一种有别于传统误差掩盖方法的,利用相临区域与未知区域的相关性来重构未知区域的方法。为了重构出有着较为复杂结构的未知区域,提出了用基于结构的方向插值来恢复结构信息周围的信息。而后,用基于纹理方向的Graph Cuts基准块排放方式来重构大块的纹理结构。为了进一步提高重建结果的质量,引入了重构顺序补偿。就像在实验结果中呈现的那样,提出的重建算法对于有着复杂结构的信息重建有着较好的作用。

While the compressed video bitstream derived from block-based predictive coding is transmitted over fading channels, lost packets could cause error propagation along motion prediction alignments. Instead of traditional error recovery mechanisms, this paper aims at reconstructing the unknown regions in lost packets utilizing inherent correlation between the lost blocks and its neighbors. For regions with complex geometric property, a structure-driven directional interpolation is proposed to restore information in the unknown region near structure. Thereafter, a directional patch placement for Graph Cuts is dedicated to achieving uniform textm＇e within unknown regions. To offer a best-first filling between structure continuity and texture confidence, a filling priority compensation is devoted to further enhancing restoration results. As demonstrated by experiment results, it could be seen that proposed restoration can achieve good restoration results even when big amount of blocks are lost in transmission.