• DocumentCode
    1322857
  • Title

    Approximate criteria for the MPEG-2 motion estimation

  • Author

    Senda, Yuzo

  • Author_Institution
    C&C Media Res. Labs., NEC Corp., Kawasaki, Japan
  • Volume
    10
  • Issue
    3
  • fYear
    2000
  • fDate
    4/1/2000 12:00:00 AM
  • Firstpage
    490
  • Lastpage
    497
  • Abstract
    This paper proposes three kinds of approximate criteria for the MPEG-2 motion estimation. They are based on the behavior of motion-compensated prediction, which uses a kind of averaging. A simplified motion estimation method is derived as a result of applying the criteria to the MPEG-2 test model motion estimator. The test model motion estimator consists of three steps: (1) finding full-pel motion vectors; (2) refining the vectors to half-pel accuracy; and (3) evaluating either the interpolative or the dual-prime prediction mode, which is a combination of the vectors. Since the second and the third steps require a huge number of data transfers, the circuit size of the steps becomes comparable to that of the first step. The simplified method solves this problem while keeping practically equivalent performance. In the second and the third steps, the simplified method reduces the number of necessary computations to less than 1%, and that of data transfers to less than 8% of the test model motion estimator. Even though the simplified method only modifies the second and third steps, the number of total data transfers, including those for the first step, is also reduced to 27% in main profile or 20% in simple profile
  • Keywords
    interpolation; motion compensation; motion estimation; prediction theory; MPEG-2 motion estimation; MPEG-2 test model motion estimator; approximate criteria; averaging; circuit size; data transfer; dual-prime prediction mode; full-pel motion vectors; half-pel accuracy; interpolative prediction mode; motion-compensated prediction; Circuits; Codecs; Interpolation; Motion estimation; Motion measurement; National electric code; Prediction methods; Predictive models; Signal to noise ratio; Testing;
  • fLanguage
    English
  • Journal_Title
    Circuits and Systems for Video Technology, IEEE Transactions on
  • Publisher
    ieee
  • ISSN
    1051-8215
  • Type

    jour

  • DOI
    10.1109/76.836294
  • Filename
    836294