A low power optimal motion search algorithm

Abstract

We propose an efficient algorithm using temporal correlation, which employs a lower computational complexity but without sacrificing the accuracy. To achieve this goal, two techniques are employed, one is recursive motion vector estimation, and the other is a hierarchical layer processing for different motion feature. With this approach, we can attain near-global optimization and local optimization. In the final evaluation, the proposed complexity is only 1/2/spl sim/1/5 of full search, but the accuracy is slightly higher than one of full search in the particular range.