Low-Complexity Heterogeneous Video Transcoding by Motion Vector Clustering

Abstract

The recently developed video compression standard, H.264/AVC surpasses the performance of previous video standards, such as MPEG-2, MPEG-4(part2), and H.263 and is therefore expected to be selected as the video standard for most digital video applications. The widely distributed infrastructure, however, continues to use the previous standards. Heterogeneous video transcoding offers a significant key to the resolution of this problem. This paper suggests a new algorithm for H.264/AVC to MPEG-2 transcoding that uses motion vector clustering to reduce the computation time with no loss of quality. Such a clustering method can reduce the number of candidate motion vectors that are gathered during the H.264 decoding stage. These candidate motion vectors consider the correlation between the direction and distance of the motion vectors in the variable blocks in H.264/AVC. The candidate motion vector that has the least distortion is then selected in the MPEG-2 encoder. The MPEG-2 encoder can therefore use the best motion vector without carrying out computations for motion estimation. The experimental results show that the proposed method can maintain a good level of video quality while reducing the computational complexity by a considerable 64%, on average, compared to a cascade transcoder.