一种自适应的运动矢量精细化算法

Adaptive motion vector refinement algorithm

针对跳帧视频转码中运动矢量合成算法可能引入较大运动矢量误差的问题,提出了一种自适应的运动矢量精细化算法,该算法根据误差大小将运动矢量分为两组,对误差较小的一组进行小范围精细化,另一组在原始运动矢量、预测运动矢量和合成运动矢量中选择最优的运动矢量,并在其周围做较大范围精细化,实验结果表明,此方法可以有效地提高图像质量,并减小输出码率。

In frame skipping transcoding,original motion vectors are used to recompose new ones for outgoing stream.However, these recomposed motion vectors may suffer larger error.In this paper we propose an adaptive motion vector refinement algorithm that classifies motion vectors in two groups.Small range refinement is applied for the group with small error.For the other group, the best motion vector is selected from original,predicted and recomposed motion vectors according to SAD （Sum of Absolute Difference）.Then larger range refinement is applied.Experimental results show that the proposed algorithm can effectively improve PSNR and reduce bit rate of the outgoing stream.