Efficient Motion Estimation Scheme for Downsizing H.264 Homogeneous Transcoding

In this paper, we propose an efficient motion estimation scheme for pixel domain downsizing homogeneous H.264 transcoding. We exploit the motion information from the original decoded frames to generate more accurate motion vector (MV) candidates and make a fast selection of the best reference frame for current macroblock (MB) in the down-scaled frame, finally based on the chosen reference frame and MV candidates, perform the mode decision. Extensive experiment results show that the rate-distortion (RD) performance of the proposed scheme is very close to the benchmark full-search method (within 0.05 dB loss and 2% bit rate increase on average) while 68% computation time is saved.