Prediction-based directional search for fast block-matching motion estimation

This paper proposes an efficient block-matching motion estimation algorithm known as prediction-based directional search (PDS). This new algorithm is applicable to a wide range of video processing applications. The algorithm uses the motion vectors in two neighboring blocks to predict a starting search point for the current block. The subsequent refining search relies on the hypothesis of monotonic block distortion surface and the center-biased characteristic of motion vector probability distribution. The cross pattern in a step and one of four possible directional rectangle search patterns in the next step are iteratively used to find the motion vector. Experiments on eleven video sequences with different characteristics shows that PDS can achieve a faster computation speed with similar or even better distortion performance compared to some existing well-known algorithms.