基于块运动轨迹相似性的H.264运动修复-错误隐藏算法  

H.264 Motion Recovery-Error Concealment Algorithm Based on Block Motion Trajectory Similarity

作　　者：南哲万[1,2] 阮秋琦[1] 安高云[1] 

机构地区：[1]北京交通大学信息科学研究所,北京100044 [2]金日成综合大学计算机学院

出　　处：《铁道学报》2013年第2期66-71,共6页Journal of the China Railway Society

基　　金：国家自然科学基金(60973060);教育部高等学校博士学科点专项科研基金(200800040008)

摘　　要：H.264视频在通过无线网络传递时,对传输错误非常敏感,由此会引起视频质量的严重下降。本文提出1种基于块运动轨迹相似性及空域统计相关性的受损块运动修复-错误隐藏算法。采用绝对运动差值总和(Sumof Absolute Difference for Motion Vector)估计运动轨迹相似性,并据此在受损宏块周围搜索可靠块;在过去帧中抽取可靠块和受损块的空域相互关系统计特征,并据此在当前帧预测受损块运动矢量;采用基于宏块分割信息的块合并来修复宏块内子块运动矢量,采用邻域插值方法恢复丢失的像素。实验结果表明,与零运动算法、宏块边界匹配算法和多项式内插法相比较,本算法执行速度较快,隐藏效果较好。H. 264 video, when being transmitted over error-prone wireless channels, may suffer from losses or errors. The lost data may result in unacceptable quality degradation. In this paper, we proposed a low complexi- ty and effective motion recovery algorithm for temporal error concealment in H. 264, which was based on block motion trajectory similarity and statistical correlation of spatial domain. The block motion trajectory similarity was estimated by using the Sum of Absolute Difference for Motion Vector （SAD-MV）, and then accordingly the available blocks for Lost Block （LB） were searched for in the lost Macro-block （MB） neighbors. The MVs of LBs in the current frame were predicted on the basis of the statisticat spatial correlation between the availa- ble blocks and LBs extracted from previous frames. The MVs of the MB blockettes were merged and recovered from the MVs of blocks subject to MB segmenting information. The lost pixels were recovered with the neigh- bourhood interpolation method. Experiments prove that the proposed method is characterized by quick execu- tion velocity and good hidden eqqed as compared with the zero MV method, MB boundary match algorithm and polynomial interpolation method.

关 键 词：错误隐藏 运动修复 视频 无线网络 H 264 

分 类 号：TN926.3[电子电信—通信与信息系统]