Parabolic motion-vector re-estimation algorithm for compressed video downscaling

For multimedia communications, there is a need to downscale a video prior to transmission because of the limitation of the bandwidth of a channel, and/or the different standards between transcoders. However, performing motion estimation in the downscaled video is computational intensive. In this paper, a parabolic motion vector re-estimation (PMVR) algorithm is proposed to predict motion vectors of the downscaled video. The proposed algorithm can significantly reduce computational complexity with slight PSNR degradation, in comparison with full search algorithm. Experimental results show that, with few additional computation, the proposed algorithm achieves a much higher quality than several existing algorithms.