A fast block-matching algorithm using smooth motion vector field adaptive search technique

Abstract

In many video standards based on inter-frame compression such as H.26x and MPEG, block-matching algorithm has been widely adopted as the method for motion estimation because of its simplicity and effectiveness. Nevertheless, since motion estimation is very complex in computing. Fast algorithm for motion estimation has always been an important and attractive topic in video compression. From the viewpoint of making motion vector field smoother, this paper proposes a new algorithm SMVFAST. On the basis of motion correlation, it predicts the starting point by neighboring motion vectors according to their SADs. Adaptive search modes are used in its search process through simply classifying motion activity. After discovering the ubiquitous ratio between the SADs of the collocated blocks in the consecutive frames, the paper proposes an effective half-stop criterion that can quickly stop the search process with good enough results. Experiments show that SMVFAST obtains almost the same results as the full search at very low computation cost, and outperforms MVFAST and PMVFAST in speed and quality, which are adopted by MPEG-4.