Exploiting Reference Frame History in H.264/AVC Motion Estimation

Motion estimation is the most crucial and time-consuming part of the H.264/AVC video compression standard. The introduction of motion search of variable block sizes in multiple reference frames has significantly increased the computational complexity. This paper proposes a fast motion estimation algorithm, most used reference first (MURF), based on the usage history of reference frames. The algorithm rearranges the search order of the reference frames based on the selection probability of the reference frames in coding the current frame. The experimental results show that the proposed algorithm, when compared to the best algorithm in H.264 reference software, achieves on average 60% reduction in search points and 52% reduction in motion estimation time with comparable video quality and negligible increase in bit-rate and memory