[1]彭宗举,邹文辉,陈芬,等.面向HEVC的时空域感知量化[J].西华师范大学学报(自然科学版),2016,37(01):67-74.[doi:10.16246/j.issn.1673-5072.2016.01.010]
 PENG Zongju,ZOU Wenhui,CHEN Fen,et al.Spatialtemporal Perception Quantization for HEVC[J].Journal of China West Normal University(Natural Sciences),2016,37(01):67-74.[doi:10.16246/j.issn.1673-5072.2016.01.010]
点击复制

面向HEVC的时空域感知量化()
分享到:

《西华师范大学学报(自然科学版)》[ISSN:1673-5072/CN:51-1699/N]

卷:
37
期数:
2016年01期
页码:
67-74
栏目:
出版日期:
2016-03-20

文章信息/Info

Title:
Spatialtemporal Perception Quantization for HEVC
作者:
彭宗举1邹文辉2陈芬1费马燕1
(1.宁波大学 信息科学与工程学院,浙江 宁波315211;2.西华师范大学 电子信息工程学院,四川 南充637009)
Author(s):
PENG Zongju1ZOU Wenhui2CHEN Fen1FEI Mayan1
(1. Faculty of Information Science and Engineering,Ningbo Zhejiang University,Ningbo Zhejiang 315211, China; 2. College of Electronic and Information Engineering,China West Normal University, Nanchong Sichuan 637009, China)
关键词:
高效视频编码感兴趣区域感知编码量化参数
Keywords:
high efficiency video codingregion of interestperceptual coding quantization parameter
分类号:
TN919
DOI:
10.16246/j.issn.1673-5072.2016.01.010
文献标志码:
A
摘要:
针对高效视频编码(High Efficiency Video Coding,HEVC)标准在编码过程中没有考虑到人类视觉系统的感知特性的问题,提出了一种基于HEVC的时空域感知量化编码策略。首先,对输入的视频序列在变换域求取恰可察觉失真(Just Noticeable Distortion,JND),在最大编码单元(Largest Coding Unit,LCU)层根据JND求取相应的量化参数。其次,通过运动估计方法对输入的视频序列求取运动区域,由于视频序列中存在着各种各样的噪声,影响运动估计的准确性,因此对运动估计后的图像进行滤波处理,以减弱噪声的影响,并将运动区域视为时域感兴趣区域,之后根据是否为运动区域分策略调节量化参数。最后,在LCU编码时,对根据空域感知特性调节的量化参数以及时域感知特性调节的量化参数进行加权,作为LCU最终编码量化参数。实验结果表明,提出的算法在编码时视觉感兴趣区域选择了细量化,而视觉非感兴趣区域选择了粗量化。感兴趣区域的PNSR相对HM11.0能提高0.20—0.59dB,率失真性能最小能提升3.8%,最大能提升6.8%。与代表性文献算法相比,PSNR提升较大。
Abstract:
To solve the drawback of high efficiency video coding (HEVC) which ignores the property of human visual system in the process of video coding, a spatialtemporal perception quantization method is proposed in this paper.Three steps were taken in the experiment. First, just noticeable distortion (JND) in transform domain is obtained for input video sequence, and then used to calculate the quantization parameter (QP) for largest coding unit (LCU). Second, motion area is extracted according to input video sequence by motion estimation. Yet due to the variety of noise in the natural sequence, motion estimation would not be accurate enough. Thus, a novel filter is proposed to weaken the influence aroused by noise. The motion area is regarded as region of interest (ROI) and used to adjust the QP. Third, the ultimate QP is obtained by weighting the QP calculated by JND and ROI. Experimental results showed that the proposed method improves the rate distortion performance and PSNR. For the same PSNR, the proposed approach can save 5.1% bit on average. Maximal bit rate saving can reach 6.8%. The PSNR in ROI can be improved by 0.20~0.59dB compared with the HEVC reference software.

备注/Memo

备注/Memo:
收稿日期:2016-01-28
基金项目:国家自然科学基金(U1301257, 61271270)
作者简介:彭宗举(1973—),四川南部人,教授,博士,博士生导师,主要从事图像处理、3D视频信号编码与通信领域的研究。
通讯作者:彭宗举,E-mail:pengzongju@nbu.edu.cn
更新日期/Last Update: 2016-03-20