Adaptive motion vector refinement for high performance transcoding

In transcoding, simple reuse of the motion vectors extracted from the incoming bit-stream may introduce severe quality degradation in many applications. In this paper we first discuss the effect of quantization error on the motion vectors. We show that due to the quantization errors, the incoming motion vectors become non-optimal. To achieve the best quality possible, a new motion estimation should be performed in the transcoder. We proposed an adaptive motion vector refinement scheme with a highly reduced computational complexity in comparison to the full-scale motion estimation. We also discuss the case when some incoming frames are dropped for the frame-rate conversion. We propose a simple and fast motion vector interpolation method. We show that the proposed adaptive refinement scheme provides video quality almost as good as that can be achieved by performing a new full-scale motion estimation but requires much less computation