Fast Motion Estimation Based on Confidence Interval
Fast Motion Estimation Based on Confidence Interval
复制标题
DOI：
10.1109/tcsvt.2014.2306035
复制
发表时间：
2014-02
影响因子：
8.4
通讯作者：
Nan Hu;E. Yang
中科院分区：
文献类型：
--
作者：
Nan Hu;E. Yang
名为高效视频编码 (HEVC) 的新视频标准最近已最终确定。与H.264/AVC视频编码标准相比，HEVC进一步提高了视频编码率失真（RD）性能，但其代价是其编码复杂度显着增加，特别是由于大块尺寸和复杂的块划分而导致其运动估计（ME）。为了降低 HEVC 中的 ME 复杂度，同时保持其 RD 性能，在本文中，我们首先将整数像素级别的 ME 公式化为统计推断问题，然后提出一种基于置信区间的 ME（CIME）方法。所提出的 CIME 方法可以应用于 HEVC 中实现的现有快速搜索之上，也可以单独应用以取代 HEVC 中实现的现有快速搜索。实验表明，对于 HEVC 的低延迟主要测试配置：1) 当应用于 HEVC 中现有的快速搜索时，所提出的 CIME 方法平均进一步减少了 70% 的整数级 ME 时间，比特率仅增加 1.0%，同时保持相同的 PSNR 重建质量；2) 当单独应用于 HEVC 中实现的现有快速搜索时，所提出的 CIME 方法实现了与 HEVC 参考软件中的快速搜索相当的性能，并且更好优于文献中最近提出的动态系统快速算法。
A new video standard called High Efficiency Video Coding (HEVC) has been recently finalized. In comparison with the H.264/AVC video coding standard, HEVC further improves the video coding rate distortion (RD) performance, but at the price of significant increase in its encoding complexity, especially in its motion estimation (ME) due to large block sizes and complicated block partition. To reduce the ME complexity in HEVC while maintaining its RD performance, in this paper we first formulate ME at the integer pixel level as a statistical inference problem and then propose a confidence interval-based ME (CIME) method. The proposed CIME method can be applied either on top of the existing fast search implemented in HEVC or on its own to replace the existing fast search implemented in HEVC. Experiments show that for the low-delay main test configuration of HEVC: 1) when applied on top of the existing fast search in HEVC, the proposed CIME method further reduces, on average, the integer-level ME time by 70% with only 1.0% increase in bit rate while maintaining the same reconstruction quality in PSNR and 2) when applied on its own to replace the existing fast search implemented in HEVC, the proposed CIME method achieves performance comparable with that of the fast search in HEVC reference software and better than that of the dynamic system fast algorithm proposed recently in the literature.