视频编码中基于时空信息运动估计的研究  

Spatial and Temporal Motion Vector Prediction and Fast Motion Estimation

作　　者：杨成军[1] 魏春城[1] 陈铭[1] 全子一[1] 

机构地区：[1]北京邮电大学图像处理与传输实验室,北京100876

出　　处：《高技术通讯》2002年第4期27-31,共5页Chinese High Technology Letters

摘　　要：提出了一种基于视频图像的时间和空间相关性信息的运动矢量预测方法。由于相邻两帧图像中的内容应具有相关性 ,图像中物体的运动轨迹也具有相似性。因此可以利用前一帧图像中宏块的运动矢量确定运动轨迹 ,进而估计当前宏块的运动矢量。但是这种运动估计是受限的 ,因为估计运动矢量时的宏块顺序是逐行的按先后顺序进行。在当前帧中 ,当前宏块预测时只能利用具有因果关系的相邻宏块 ,但其与相邻宏块并非都具有因果关系 ,因此给出了预测矢量的资格检验方法 ,用以判断和剔除伪预测矢量。在搜索过程中采用菱形搜索 (Pyramid) ,相邻的两次搜索过程 ,要去除两个重叠区域。上述算法减少了计算量 ,对Suzie序列测试结果 ,平均搜索次数是 14～ 17次 ,并且解码图像的信噪比PSNR与FS搜索相比降级不多。A novel method of motion vector prediction based on the spatial-temporal relativity has been described in this paper. Because the contents of adjacent frames should have similarity, so do the motion trajectory. So the motion trajectory from the pre-frame is figured, then using it to predicate a motion vector of current MacroBlock. But the prediction is limited, because the causality MacroBlocks(left and above current MacroBlock) are only adopted, so the trajectory should be qualified. The authors adopt a criterion to cancel the pseudo part of trajectory. During the motion vector search, the pyramid search is adapted, and the position overlapped of two continues search is skipped. The average positions of search are 14～17 and the PSNR of decoded image is near the result of FS.

关 键 词：时空信息 视频编码 运动估计 块区配 运动矢量预测 时空相关性 视频压缩 

分 类 号：TN919.81[电子电信—通信与信息系统]