A Novel Macro-Block Group Based AVS Coding Scheme for Many-Core Processor

  • Authors:

  • Zhenyu Wang;Luhong Liang;Xianguo Zhang;Jun Sun;Debin Zhao;Wen Gao

  • Affiliations:

  • Peking University, Beijing, P.R. China 100871;Key Lab of Intelligent Information Processing, Chinese Academy of Sciences (CAS), Beijing, China 100190 and Institute of Computing Technology, CAS, Beijing, China 100190;Peking University, Beijing, P.R. China 100871;Peking University, Beijing, P.R. China 100871;Harbin Institute of Technology, Harbin, P.R. China 150001;Peking University, Beijing, P.R. China 100871

  • Venue:
  • PCM '09 Proceedings of the 10th Pacific Rim Conference on Multimedia: Advances in Multimedia Information Processing
  • Year:
  • 2009

Quantified Score

Hi-index 0.00

Visualization

Abstract

The slice-level parallelism is popular in parallel video coding. However, the quality loses greatly because the dependency between macro-blocks is broken, especially on many-core platforms. To address this problem, a novel Macro-Block Group (MBG) decomposition scheme is presented for parallel AVS coding. In the proposed scheme, video frames are equally divided into rectangular MBG regions, each consists of more rows and less columns than the slice-level scheme. Since MBG is not supported by AVS, a vertical partitioning scheme is introduced, and the mode confining and MVD adjusting techniques are utilized to keep consistency with the standard. In practice, our parallel encoder is developed on the TILE64 platform, where P/B frames use the MBG-level parallelism and I frames use the macro-block-level parallelism. Experiments show that the proposed scheme can achieve a reduction of 52% (IPPP) and 41% (IBBP) in quality loss while keeping the same speed-up compared with the slice-level parallelism.