多层次细粒度并行HEVC帧内模式选择算法

发布时间：2018-03-07 23:26

本文选题：高效视频编码　切入点：帧内预测　出处：《计算机研究与发展》2016年04期 　论文类型：期刊论文

【摘要】：在众核平台上并行加速是解决高效视频编码(high efficiency video coding,HEVC)标准编码复杂度高的有效方法.传统的粗粒度并行方案如Tiles和WPP未能在并行度和编码质量之间取得较好的平衡,对编码质量影响较大或者并行度不高.充分挖掘HEVC帧内模式选择中的并行性,提出了一种在CTU内使用的多层次细粒度的帧内模式选择算法.具体说来,对帧内模式选择过程进行了子任务划分,分析并消除了相邻编码块之间多种阻碍并行计算的数据依赖关系,包括帧内预测参考像素依赖、预测模式依赖和熵编码依赖等,实现了同一个CTU内所有层次的细粒度编码块的代价计算和模式选择并行进行.将算法在Tile-Gx36平台上实现,实验结果表明此并行算法与HEVC参考代码HM相比能获得18倍的整体编码加速比而且编码质量损失较小(码率上升3%).
[Abstract]:Parallel acceleration on multi-kernel platform is an effective method to solve the high complexity of high efficiency video coding standard. Traditional coarse-grained parallel schemes such as Tiles and WPP can not achieve a good balance between parallelism and coding quality. This paper fully exploits the parallelism in HEVC intra mode selection, and proposes a multi-level fine-grained intra-frame pattern selection algorithm used in CTU. The intra-frame mode selection process is subdivided, and several data dependencies that hinder parallel computation between adjacent coding blocks are analyzed and eliminated, including intra prediction reference pixel dependency, prediction mode dependence and entropy coding dependency, etc. The cost calculation and pattern selection of all levels of fine-grained coding blocks in the same CTU are implemented in parallel. The algorithm is implemented on the Tile-Gx36 platform. Experimental results show that compared with HEVC reference code HM, the parallel algorithm can obtain 18 times global coding speedup and lower coding quality loss (code rate increases by 3%).
【作者单位】：中国科学院智能信息处理重点实验室(中国科学院计算技术研究所);中国科学院大学;
【基金】：国家自然科学基金项目(61379084;61402440) 中国科学院科研装备研制项目(YZ201321)~~
【分类号】：TN919.81

【相似文献】