A study on motion compensated prediction in DCT domain with multiple mode selection

Kazuhisa Yamagisi, Jiro Katto, Yasuhiko Yasuda

    Research output: Chapter in Book/Report/Conference proceedingConference contribution

    Abstract

    In this paper, we propose a motion vector search algorithm that employs the Sum of Absolute Difference (SAD) minimization in DCT domain as a distortion measure. In addition, block based motion estimation is carried out with multiple modes in this frequency domain, and a mode is selected according to the Lagrangian cost function [1, 2, 3]. Four modes are designed. The first two modes are a mode with 1 motion vector (INTER1V) and a mode with 4 motion vectors (INTER4V) per macroblock that are similar to the H.263 [4]. The last two modes use 4 motion vectors per macroblock (INTER1V+) or per block (INTER4V+), where the 8×8 DCT coefficients are firstly separated into 4 groups and a motion vector is independently assigned to each group. The average SNR (Signal to Noise Ratio) gains between 0.3 and 1.0dB are obtained for tested sequences when compared with TMN. Our proposal method provides true motion vectors for low frequency and minimizes distortion for higher frequency. Therefore our proposed method also reduces subjective visual degradation.

    Original languageEnglish
    Title of host publicationPicture Coding Symposium
    Pages87-91
    Number of pages5
    Publication statusPublished - 2003
    EventPicture Coding Symposium - Saint Malo
    Duration: 2003 Apr 232003 Apr 25

    Other

    OtherPicture Coding Symposium
    CitySaint Malo
    Period03/4/2303/4/25

    Fingerprint

    Motion estimation
    Cost functions
    Signal to noise ratio
    Degradation

    ASJC Scopus subject areas

    • Computer Graphics and Computer-Aided Design
    • Signal Processing

    Cite this

    Yamagisi, K., Katto, J., & Yasuda, Y. (2003). A study on motion compensated prediction in DCT domain with multiple mode selection. In Picture Coding Symposium (pp. 87-91)

    A study on motion compensated prediction in DCT domain with multiple mode selection. / Yamagisi, Kazuhisa; Katto, Jiro; Yasuda, Yasuhiko.

    Picture Coding Symposium. 2003. p. 87-91.

    Research output: Chapter in Book/Report/Conference proceedingConference contribution

    Yamagisi, K, Katto, J & Yasuda, Y 2003, A study on motion compensated prediction in DCT domain with multiple mode selection. in Picture Coding Symposium. pp. 87-91, Picture Coding Symposium, Saint Malo, 03/4/23.
    Yamagisi K, Katto J, Yasuda Y. A study on motion compensated prediction in DCT domain with multiple mode selection. In Picture Coding Symposium. 2003. p. 87-91
    Yamagisi, Kazuhisa ; Katto, Jiro ; Yasuda, Yasuhiko. / A study on motion compensated prediction in DCT domain with multiple mode selection. Picture Coding Symposium. 2003. pp. 87-91
    @inproceedings{27c90f68e347415eb408047102ce6402,
    title = "A study on motion compensated prediction in DCT domain with multiple mode selection",
    abstract = "In this paper, we propose a motion vector search algorithm that employs the Sum of Absolute Difference (SAD) minimization in DCT domain as a distortion measure. In addition, block based motion estimation is carried out with multiple modes in this frequency domain, and a mode is selected according to the Lagrangian cost function [1, 2, 3]. Four modes are designed. The first two modes are a mode with 1 motion vector (INTER1V) and a mode with 4 motion vectors (INTER4V) per macroblock that are similar to the H.263 [4]. The last two modes use 4 motion vectors per macroblock (INTER1V+) or per block (INTER4V+), where the 8×8 DCT coefficients are firstly separated into 4 groups and a motion vector is independently assigned to each group. The average SNR (Signal to Noise Ratio) gains between 0.3 and 1.0dB are obtained for tested sequences when compared with TMN. Our proposal method provides true motion vectors for low frequency and minimizes distortion for higher frequency. Therefore our proposed method also reduces subjective visual degradation.",
    author = "Kazuhisa Yamagisi and Jiro Katto and Yasuhiko Yasuda",
    year = "2003",
    language = "English",
    isbn = "2726112420",
    pages = "87--91",
    booktitle = "Picture Coding Symposium",

    }

    TY - GEN

    T1 - A study on motion compensated prediction in DCT domain with multiple mode selection

    AU - Yamagisi, Kazuhisa

    AU - Katto, Jiro

    AU - Yasuda, Yasuhiko

    PY - 2003

    Y1 - 2003

    N2 - In this paper, we propose a motion vector search algorithm that employs the Sum of Absolute Difference (SAD) minimization in DCT domain as a distortion measure. In addition, block based motion estimation is carried out with multiple modes in this frequency domain, and a mode is selected according to the Lagrangian cost function [1, 2, 3]. Four modes are designed. The first two modes are a mode with 1 motion vector (INTER1V) and a mode with 4 motion vectors (INTER4V) per macroblock that are similar to the H.263 [4]. The last two modes use 4 motion vectors per macroblock (INTER1V+) or per block (INTER4V+), where the 8×8 DCT coefficients are firstly separated into 4 groups and a motion vector is independently assigned to each group. The average SNR (Signal to Noise Ratio) gains between 0.3 and 1.0dB are obtained for tested sequences when compared with TMN. Our proposal method provides true motion vectors for low frequency and minimizes distortion for higher frequency. Therefore our proposed method also reduces subjective visual degradation.

    AB - In this paper, we propose a motion vector search algorithm that employs the Sum of Absolute Difference (SAD) minimization in DCT domain as a distortion measure. In addition, block based motion estimation is carried out with multiple modes in this frequency domain, and a mode is selected according to the Lagrangian cost function [1, 2, 3]. Four modes are designed. The first two modes are a mode with 1 motion vector (INTER1V) and a mode with 4 motion vectors (INTER4V) per macroblock that are similar to the H.263 [4]. The last two modes use 4 motion vectors per macroblock (INTER1V+) or per block (INTER4V+), where the 8×8 DCT coefficients are firstly separated into 4 groups and a motion vector is independently assigned to each group. The average SNR (Signal to Noise Ratio) gains between 0.3 and 1.0dB are obtained for tested sequences when compared with TMN. Our proposal method provides true motion vectors for low frequency and minimizes distortion for higher frequency. Therefore our proposed method also reduces subjective visual degradation.

    UR - http://www.scopus.com/inward/record.url?scp=0344467067&partnerID=8YFLogxK

    UR - http://www.scopus.com/inward/citedby.url?scp=0344467067&partnerID=8YFLogxK

    M3 - Conference contribution

    SN - 2726112420

    SP - 87

    EP - 91

    BT - Picture Coding Symposium

    ER -