• DocumentCode
    597897
  • Title

    New chroma intra prediction modes based on linear model for HEVC

  • Author

    Xingyu Zhang ; Au, Oscar C. ; Jingjing Dai ; Chao Pang ; Feng Zou

  • Author_Institution
    Dept. of Electron. & Comput. Eng., Hong Kong Univ. of Sci. & Technol., Hong Kong, China
  • fYear
    2012
  • fDate
    Sept. 30 2012-Oct. 3 2012
  • Firstpage
    197
  • Lastpage
    200
  • Abstract
    RGB-to-YUV conversion reduces the inter-channel redundancy in video compression. However, inter-channel correlation is still observed locally. In HEVC HM4.0, LM mode uses a linear model to predict chroma from luma as a chroma intra prediction mode. The linearity is derived from the reconstructed causal pixels. Specifically, for a NxN chroma block, the above N neighbors and the left N neighbors are used for derivation. Considering the variety of the content inside a block, the linearity derived in LM mode does not always match the linearity of the current block. In this paper, we propose two new modes LML and LMA for chroma intra prediction. Those two modes are quite similar to LM mode, except that they use different neighbors for linearity derivation. By adding the proposed modes, about an average of 0.2%, 5.9%, 6.7% BD-rate reduction is achieved under all intra configuration for Y, Cb and Cr components respectively.
  • Keywords
    data compression; image colour analysis; image reconstruction; image resolution; image sequences; video coding; HEVC HM4.0; LM mode; LMA mode; LML mode; RGB-to-YUV conversion; causal pixel reconstruction; chroma block; chroma intra prediction modes; interchannel redundancy reduction; linear model; video compression; video sequences; Correlation; Decoding; Equations; Linearity; Mathematical model; Optimization; Predictive models; Chroma intra prediction; Linear model; video compression;
  • fLanguage
    English
  • Publisher
    ieee
  • Conference_Titel
    Image Processing (ICIP), 2012 19th IEEE International Conference on
  • Conference_Location
    Orlando, FL
  • ISSN
    1522-4880
  • Print_ISBN
    978-1-4673-2534-9
  • Electronic_ISBN
    1522-4880
  • Type

    conf

  • DOI
    10.1109/ICIP.2012.6466829
  • Filename
    6466829