EURASIP Journal on Advances in Signal Processing 
Volume 2007 (2007), Article ID 75310, 12 pages
doi:10.1155/2007/75310
Research Article

Efficient MPEG-2 to H.264/AVC Transcoding of Intra-Coded Video

Jun Xin,1 Anthony Vetro,2 Huifang Sun,2 and Yeping Su3

 1Xilient Inc., 10181 Bubb Road, Cupertino 95014, CA, USA
2Mitsubishi Electric Research Labs, 201 Broadway, Cambridge 02139, MA, USA
3Sharp Labs of America, 5750 NW Pacific Rim Boulevard, Camas 98607, WA, USA
Received 3 October 2006; Revised 30 January 2007; Accepted 25 March 2007

Recommended by Yap-Peng Tan

Abstract

This paper presents an efficient transform-domain architecture and corresponding mode decision algorithms for transcoding intra-coded video from MPEG-2 to H.264/AVC. Low complexity is achieved in several ways. First, our architecture employs direct conversion of the transform coefficients, which eliminates the need for the inverse discrete cosine transform (DCT) and forward H.264/AVC transform. Then, within this transform-domain architecture, we perform macroblock-based mode decisions based on H.264/AVC transform coefficients, which is possible using a novel method of calculating distortion in the transform domain. The proposed method for distortion calculation could be used to make rate-distortion optimized mode decisions with lower complexity. Compared to the pixel-domain architecture with rate-distortion optimized mode decision, simulation results show that there is a negligible loss in quality incurred by the direct conversion of transform coefficients and the proposed transform-domain mode decision algorithms, while complexity is significantly reduced. To further reduce the complexity, we also propose two fast mode decision algorithms. The first algorithm ranks modes based on a simple cost function in the transform domain, then computes the rate-distortion optimal mode from a reduced set of ranked modes. The second algorithm exploits temporal correlations in the mode decision between temporally adjacent frames. Simulation results show that these algorithms provide additional computational savings over the proposed transform-domain architecture while maintaining virtually the same coding efficiency.