Fast Motion Estimation Scheme for H.264/AVC Using Adaptive Single and Multiple Reference Picture Selections

Abstract In video coding standard of H.264/AVC, motion estimation (ME) using multiple reference pictures improves compression efficiency considerably. However, the motion estimation includes a large amount of useless computations regardless of picture contents. This paper proposes a fast motion estimation algorithm that removes the worthless computations in the motion estimation process with switching single and multiple reference pictures adaptively. The proposed algorithm classifies blocks into valid and invalid blocks for the multiple references based ME, and removes the unnecessary computations by applying a single reference based ME to the invalid blocks. In order to evaluate the performance of the proposed algorithm, measures with respects to image quality, bit rate, and motion estimation time are compared with those of the scheme in a reference software JM 9.5. The simulation results show that the proposed algorithm can considerably save 39% on average motion estimation time while keeping the image quality (0.025 dB degradation) and the bit rate (0.8% decrease) as good as the reference algorithm.