Reducing Motion Estimation Complexity in MPEG-2 TO H.264 Transcoding

The two key problems in video transcoding are complexity reduction and quality management. The complexity of H.264 encoding makes complexity reduction even more important. A good transcoder design reduces complexity with negligible loss in quality. While motion estimation complexity can be easily reduced by reducing the search range or eliminating the coding modes, these approaches will cause significant quality drop. This paper presents reduced complexity motion estimation for MPEG-2 to H.264 transcoding. The proposed approach uses MPEG-2 motion vectors to dynamically reduce the search range and the motion vector refinement window. This approach is based on the hypothesis that longer MPEG-2 motion vectors indicate higher motion and hence a larger search window as well as a larger refinement window is necessary. Experimental results show that the proposed approach reduces the motion estimation complexity by as much as 85% with negligible loss in PSNR. We show that the results hold across video resolutions and bitrates.