文章摘要
引用本文:陈清坤,施隆照,黄 博,高小虹.面向HEVC的运动估计快速算法和硬件架构[J].福州大学学报(自然科学版),2018,46(5):636~643
面向HEVC的运动估计快速算法和硬件架构
Fast algorithm and hardware architecture for HEVC motion estimation
  
DOI:10.7631/issn.1000-2243.17333
中文关键词: 运动估计  硬件架构  高效视频编码标准  帧间预测
英文关键词: motion estimation  hardware architecture  high efficiency video coding  inter prediction
基金项目:
作者单位
陈清坤 福州大学物理与信息工程学院福建 福州 350116 
施隆照 福州大学物理与信息工程学院福建 福州 350116 
黄 博 福州大学物理与信息工程学院福建 福州 350116 
高小虹 福州大学物理与信息工程学院福建 福州 350116 
摘要点击次数: 68
全文下载次数: 49
中文摘要:
      提出确定共享搜索区域的方法,实现硬件搜索区域共享,能有效提高数据复用和减少参考像素带宽. 进而,提出改进的钻石搜索算法,该算法考虑了硬件资源消耗和预测单元大小,使预测单元自适应选择搜索模板. 最后,基于改进的钻石搜索算法提出新的硬件架构,该架构通过灵活选择不同处理单元(PE)数目,实现两种基本处理单元,使不同尺寸预测单元都有较高的处理速度和硬件资源利用率. 算法仿真结果表明,本算法与参考代码HM16.7相比较,编码性能损失可忽略不计,但更适合于硬件实现. 用Altera的 Stratix IV系列芯片在QUARTUS II中综合结果表明,周期数比现有文献更少,本架构最大工作频率可达到317.56MHz,并且实现1080px @23.7帧·s-1的吞吐率.
英文摘要:
      This paper proposes a method to share a fixed hardware search region which can reuse data effectively and reduce bandwidth of the reference pixel at the same time. Secondly,an improved diamond search algorithm is proposed. The algorithm considers the hardware resource consumption and the size of the prediction unit. And the prediction unit adaptively selects the search template. Finally,this paper further proposes a new hardware architecture based on improved diamond search algorithm. This hardware architecture achieves two basic process units by flexibly selecting the number of process element ( PE ). So that different sizes of prediction units have high processing speed and hardware resource utilization. The proposed algorithm is more suitable for hardware implementation with negligible loss compared with the reference code HM16.7. The architecture has been synthesized for an Altera Stratix IV FPGA in QUARTUS II. The architecture has less the number of cycles in comparison to related works. For the architecture reaches a maximum frequency of 317.56MHz,it is possible to achieve 1 080px @23.7f·s-1.
查看全文   查看/发表评论  下载PDF阅读器
关闭