Optimization of the three-step search algorithm by exclusion of stationary macroblocks from the search process

A typical video frame contains many macroblocks that experience little motion with respect to the reference frame. These macroblocks are referred to as stationary macroblocks. This paper aims at accelerating the three-step search algorithm for block motion estimation (BME) by excluding these stationary macroblocks from the search process. A method to detect these stationary macroblocks by comparing the mean absolute difference between the current macroblock and the reference macroblock with a given threshold is first proposed. In order to make the algorithm more efficient, a mechanism to adapt the threshold level to the changing statistics of the video sequence is then developed. Simulation results demonstrate that by using the proposed algorithm, a speedup of 20 to 40 percent is achieved for typical video test sequences. The accuracy of the proposed algorithm remains about the same as that of the three-step search algorithm.