• DocumentCode
    2340342
  • Title

    New Prediction Modes for Parallel Processing of H.264/AVC

  • Author

    Bando, Takafumi ; Hirai, Naoyuki ; Song, Tian ; Shimamoto, Takashi

  • Author_Institution
    Grad. Sch. of Eng., Univ. of Tokushima, Tokushima, Japan
  • Volume
    1
  • fYear
    2011
  • fDate
    14-15 May 2011
  • Firstpage
    344
  • Lastpage
    347
  • Abstract
    New coding algorithm is required to be compatible to the improvement of the new trend of hardware devices. This work proposes a novel parallel processing algorithm for H.264/AVC which is dedicated to GPU implementation. Firstly, the original pixel data is used in the rate-distortion optimization (RDO) process instead of the reconstructed pixel data to realize the complete MB-level parallel processing of INTRA coding. It can help to parallel encode all the MB in one frame in the same time on GPU. Next, a new temporal prediction concept is introduced in the traditional spatial prediction to compensate the weak point of individual usage of spatial or temporal prediction. The new algorithm uses the selected block in the previous frame as the reference data to perform spatial prediction. The simulation results show that when using the original pixel in the RDO process it will induce slight PSNR loss. However, when the proposed new modes are introduced the RD-curve shows about ldB improvement than the original H.264/AVC.
  • Keywords
    computer graphic equipment; coprocessors; optimisation; parallel processing; video coding; GPU implementation; H.264-AVC; INTRA coding; MB level parallel processing; PSNR loss; RD-curve; coding algorithm; rate distortion optimization process; temporal prediction concept; Encoding; Graphics processing unit; Parallel processing; Pixel; Prediction algorithms; Redundancy; Simulation; GPU; H.264/AVC; INTER; INTRA;
  • fLanguage
    English
  • Publisher
    ieee
  • Conference_Titel
    Multimedia and Signal Processing (CMSP), 2011 International Conference on
  • Conference_Location
    Guilin, Guangxi
  • Print_ISBN
    978-1-61284-314-8
  • Electronic_ISBN
    978-1-61284-314-8
  • Type

    conf

  • DOI
    10.1109/CMSP.2011.76
  • Filename
    5957437