Fast motion estimation algorithm using adaptive motion vector field search

Motion estimation is rather complex in computation and spends much time in the inter-frame video compression. Fast motion estimation has always been a favorite to the researchers in the field. From the viewpoint of making motion vector field more homogenous, this paper proposed a novel fast algorithm, which is composed of the prediction of initial search point, adaptive search modes for the video sequences with different motion, and effective half-stop criteria. Experiments show that the algorithm could obtain almost the same PSNR as the full search at very low computation cost, and outperforms both MVFAST and PMVFAST, which were adopted by the current MPEG-4, in terms of speed and quality.