Reusable Architecture and Complexity-Controllable Algorithm for the Integer/Fractional Motion Estimation of H.264

Motion estimation is the most computational intensive part of H.264 video coding. The motion estimation of H.264 includes integer motion estimation and fractional motion estimation. In this paper, a reusable architecture for both of them is proposed. This architecture can support fractional motion estimation and fast integer motion estimation algorithms including diamond search and cross search. The complexity of the motion estimation algorithm implemented on the proposed architecture can be flexibly controlled to meet variable system requirement. Cycle allocation is used in the algorithm design so that the cycle budget can be fully utilized. Experimental results show that this architecture-algorithm co-design can support HD (1280times720) video real-time encoding at 30 frames per second with 238K logic gates and 117 Mhz operation frequency.