Abstract:

This paper proposes a fast algorithm based on motion classification and direction prediction in which the spatialtemporal correlations of adjacent blocks’ motion vectors are used. The algorithm predicts the motion classification and relative motion of blocks and adaptively chooses different search method with motion classification information of blocks. For relative fast motion blocks, it predicts the motion direction and chooses corresponding search patterns. Experiments show that the algorithm can save more search time than other traditional algorithms with the same accuracy.

Key words: motion vector;motion estimation;motion classification;direction prediction