A complexity adjustable algorithm for motion estimation in H.264

To deploy highly efficient video coding technologies on different hardware platform, complexity adjustable algorithms are required. One of the significant resource consumers is motion estimation (ME). This paper proposes a complexity adjustable algorithm for ME to remedy this issue. First, we build a mathematic model to indicate that macroblocks (MBs) with intensive motion are worthier of finer searching than other MBs. And then, under a specified computational restriction, a principal is delivered to determine what kinds of MBs should be thoroughly searched prior to others. According to our experimental results, the proposed algorithm not only freely adjusts the complexity for ME but also attains much higher speed without any coding loss than using fast ME algorithms along. Our algorithm, with complexity scalability, can be easily used on platforms with varied computational capabilities. Also, the algorithm allow us dynamically choose different ME strategies to meet the need for the optimal balance between distortion gain and computational complexity.