欢迎访问中国科学院大学学报,今天是

中国科学院大学学报 ›› 2013, Vol. 30 ›› Issue (6): 813-818.DOI: 10.7523/j.issn.2095-6134.2013.06.015

• 信息与电子科学 • 上一篇    下一篇

细粒度并行归一化部分失真运动估计

袁竞杰1, 张清毅1, 马宜科2, 宋风龙2   

  1. 1. 中国矿业大学(北京)机电与信息工程学院, 北京 100083;
    2. 中国科学院计算技术研究所计算机体系结构国家重点实验室, 北京 100190
  • 收稿日期:2012-12-19 修回日期:2013-03-29 发布日期:2013-11-15
  • 通讯作者: 袁竞杰
  • 基金资助:

    国家自然科学基金(61100013)和中央高校基本科研业务费专项(2010YJ19)资助

Fine-grained parallel algorithm for normalized partial distortion search

YUAN Jing-Jie1, ZHANG Qing-Yi1, MA Yi-Ke2, SONG Feng-Long2   

  1. 1. College of Mechanical and Electrical Engineering, China University of Mining & Technology, Beijing 100083, China;
    2. State Key Laboratory of Computer Architecture, Institute of Computing Technology, Chinese Academy of Sciences, Beijing 100190, China
  • Received:2012-12-19 Revised:2013-03-29 Published:2013-11-15

摘要:

移动视频编码应用对实时性要求越来越高,传统编码器中使用的串行运动估计算法难以满足实时编码要求.本文并行化移动编码中典型的运动估计算法——归一化部分失真搜索.采用比帧和宏块更小的候选块作为并行粒度,保持归一化部分失真快速排除非最佳候选块优势,同时充分利用多核计算资源.4核CPU平台上实验结果表明,相比串行算法,该并行算法在计算量增加不超过1.2%的前提下,实现了3.88至3.96的加速比.

关键词: 视频编码, 运动估计, 部分失真搜索, 并行计算

Abstract:

Serial motion estimation does not satisfy real-time requirements of video coding in mobile terminal. A divided-spiral-path parallel motion estimation based on normalized partial distortion search is proposed to make full use of multi-core computing power, and it retains low computational complexity of the serial algorithm. Experimental results show that the proposed algorithm achieves speedup ratios of 3.88-3.96 on a 4-core CPU platform while the cost of trivial computational effort increases by less than 1.2%.

Key words: video coding, motion estimation, partial distortion search, parallel computing

中图分类号: