A fast motion estimation using a three-dimensional reference motion vector

Based on a study of the motion vector (MV) obtained by the Full Search algorithm among several test sequences, we found that the prediction process for motion estimation in many existing algorithms are not optimal. We optimize the prediction process by classifying the blocks into four different classes based on a newly proposed three-dimensional reference MV. Different search patterns and strategies are used for each class. The proposed algorithm can improve computation requirement as high as 43.9% than the diamond search algorithm.