基于邻域运动一致性的运动矢量快速构建算法

Motion vector composition algorithm based on the coherence of adjacent movement

在时移电视应用中,视频快进播放要求视频服务器能够通过跳帧抽取视频序列中的I帧和P帧的方式输出快速播放码流,当所跳帧中含有P帧时,必需对抽取出来的快进码流中的P帧进行重编码。为提高抽取后重编码效率,可利用码流中的原始运动矢量对跳帧后新参考帧条件下的运动矢量进行预测。在FDVS和ADVS算法基础上,提出了一种基于宏块及其相邻宏块运动一致性的算法来修正原始运动矢量,并通过实验将其与现有的FDVS和ADVS算法进行比较。

In the application of video on demand, video stream could be tanscoded with every I frame remained and P frame selectively abstracted to realize fast broadcasting. While receding predict frames whose reference frame is dropped in the transcoder to realize fast broadcasting, a motion vector composition algorithm by reusing motion vectors from the compressed video stream is used to reduce the computational complexity. This paper proposes an algorithm to composite the motion vector according to the motion mode relativity between the target macroblock and its adjacent macroblocks. The validity of the proposed algorithm is discussed by analyzing the statistics and the comparison to FDVS and ADVS algorithms.