Real-time MPEG-2 to H.264/AVC Transcoding



In this chapter, a significantly fast MPEG-2 to H.264/AVC transcoder will be introduced. The proposed efficient transcoder takes advantage of the information stored in the MPEG-2 bit-stream, i.e. the DCT coefficients, to accelerate the Intra prediction process in the H.264/AVC side of the transcoder. While MPEG-2 side of the transcoder does not employ Intra prediction method, the H.264/AVC side eliminates spatial redundancy by the mean of the Intra frame prediction technique. Hence, both the mode selection process and the direction prediction process in the H.264/AVC encoder are responsible for a non-negligible portion of the processing delay in the transcoder. For that reason, we propose a dramatically fast Intra mode selection and direction prediction algorithm, which has been designed and optimized specially for real-time MPEG-2 to H.264/AVC transcoders that are targeting to be used in mobile applications. The proposed MPEG-2 to H.264/AVC transcoding solution proposes an early Intra mode decision and direction prediction techniques, which is based on the spatial activity analysis using our novel smoothness determining factor and our optimized edge orientation detector. Moreover, the proposed fast Intra frame prediction algorithm utilizes the DCT coefficients directly from the MPEG-2 side of the transcoder to eliminate the possible redundant operations and efficiently produce the Intra prediction mode and the macro-block reconstruction direction for the H.264/AVC side of the transcoder. The Experimental results, using the H.264/AVC standard benchmark frames, show 92 % transcoding run-time elimination at the cost of a slight PSNR degradation (less than 3.5 %) compared with the standard conventional cascaded transcoder.

Copyright information

© Springer International Publishing Switzerland 2014

Authors and Affiliations

  • Tarek Elarabi
    • 1
  • Ahmed Abdelgawad
    • 2
  • Magdy Bayoumi
    • 3
  1. 1.IntelSanta ClaraUSA
  2. 2.Central Michigan UniversityMount PleasantUSA
  3. 3.University of Louisiana at LafayetteLafayetteUSA

Personalised recommendations