基于学习模型的3D-HEVC提前Merge模式终止算法

doi:10.11959/j.issn.1000-436x.2019165

通信学报 ›› 2019, Vol. 40 ›› Issue (7): 104-113.doi: 10.11959/j.issn.1000-436x.2019165

基于学习模型的3D-HEVC提前Merge模式终止算法

李跃^1,²,杨高波²(),丁湘陵³,朱亚培⁴

¹ 南华大学计算机学院，湖南衡阳 421001
² 湖南大学信息科学与工程学院，湖南长沙 410082
³ 湖南科技大学计算机科学与工程学院，湖南湘潭 411201
⁴ 衡阳师范学院物理与电子学院，湖南衡阳 421002

修回日期:2019-06-27 出版日期:2019-07-25 发布日期:2019-07-30
作者简介:李跃（1988- ），男，湖南衡阳人，博士，南华大学讲师，主要研究方向为 HEVC、3D-HEVC、VVC编码优化等。|杨高波（1974- ），男，湖南岳阳人，博士，湖南大学教授、博士生导师，主要研究方向为图像/视频信号处理、多媒体通信、数字媒体内容安全等。|丁湘陵（1981- ），男，湖南株洲人，博士，湖南科技大学副教授，主要研究方向为视频信息安全、图像处理等。|朱亚培（1988- ），女，河南驻马店人，衡阳师范学院讲师，主要研究方向为图像/视频压缩等。
基金资助:
国家重点研发计划基金资助项目(2018YFB1003205);国家自然科学基金资助项目(61572183)

Early Merge mode decision algorithm for 3D-HEVC based on learning model

Yue LI^1,²,Gaobo YANG²(),Xiangling DING³,Yapei ZHU⁴

¹ Computer School,University of South China,Hengyang 421001,China
² School of Computer Science and Electronic Engineering,Hunan University,Changsha 410082,China
³ School of Computer Science and Engineering,Hunan University of Science and Technology,Xiangtan 411201,China
⁴ Faculty of Physics and Electronic Information Science,Hengyang Normal University,Hengyang 421002,China

Revised:2019-06-27 Online:2019-07-25 Published:2019-07-30
Supported by:
The National Key R ＆ D Program of China(2018YFB1003205);The National Natural Science Foundation of China(61572183)

摘要/Abstract

摘要：

作为高效视频编码（HEVC）的扩展，3D-HEVC标准有效地提高了3D视频的压缩效率，但是也带来了很高的编码计算复杂度。为了显著地降低3D-HEVC编码复杂度，提出了一种提前Merge模式终止算法。首先，提取 Merge 模式编码后的残差信号作为特征信息；然后，根据当前编码帧内已经编码的编码单元（CU）的最优Merge模式残差信号建立学习模型；最后，提取当前CU的Merge模式的残差信号，并且利用学习模型预测Merge模式是否为最优模式。实验结果表明，提出的提前Merge模式终止算法分别将3D-HEVC纹理视点和深度图编码的时间降低了41.9%和24.3%，且编码性能的降低几乎可忽略不计。相较于现有的提前Merge模式算法，提出的提前Merge模式终止算法能进一步降低3D-HEVC的编码时间，并且设计简单，易于集成到3D-HEVC测试模型。

关键词: 3D高效视频编码, 快速Merge模式决定, 残差信号, 学习模型

Abstract:

3D-high efficiency video coding (3D-HEVC) standard is an extension of HEVC.Though 3D-HEVC effectively improves the compression efficiency of 3D video,it also brings huge computational complexity.To greatly reduce the 3D-HEVC coding complexity,an early Merge mode decision approach was proposed.The residual signal that encoded by the Merge mode was firstly extracted as feature information.A learning model was established in terms of the residual signals of the coding unit (CU) in current frame that used early Merge mode as the optimal mode.Finally,the residual signal was extracted for the Merge mode of current CU,and the learning model was used to predict whether the Merge mode was the optimal mode or not.Experimental results show that the proposed early Merge mode decision approach reduces the coding times of 3D-HEVC texture views and depth maps about 41.9% and 24.3%,respectively,and the coding performance degradation is almost negligible.Compared with existing early Merge mode decision approaches,the proposed approach further reduces the coding time,and can be easily integrated into the 3D-HEVC test model due to its design simplicity.

Key words: 3D-high efficiency video coding, early Merge mode decision, residual signal, learning model

中图分类号:

TN919.8

李跃,杨高波,丁湘陵,朱亚培. 基于学习模型的3D-HEVC提前Merge模式终止算法[J]. 通信学报, 2019, 40(7): 104-113.

Yue LI,Gaobo YANG,Xiangling DING,Yapei ZHU. Early Merge mode decision algorithm for 3D-HEVC based on learning model[J]. Journal on Communications, 2019, 40(7): 104-113.

图/表 11

图1

图2

图3

图5

图4

表1

表2

表3

表4

图6

表5

参考文献 28

[1]	陈亮, 顾翔, 王进 . 多速率多节点 IEEE 802.11 网络的视频传输策略[J]. 通信学报, 2017,38(3): 34-44.
	CHEN L , GU X , WANG J. . Video transmission strategy in multi-rate and multi-node IEEE 802.11 networks[J]. Journal on Communications, 2017,38(3): 34-44.
[2]	SULLIVAN G J , OHM J , HAN W ,et al. Overview of the high efficiency video coding (HEVC) standard[J]. IEEE Transactions on Circuits and Systems for Video Technology, 2012,22(12): 1649-1668.
[3]	MULLER K , SCHWARZ H , MARPE D ,et al. 3D high-efficiency video coding for multi-view video and depth data[J]. IEEE Transactions on Image Processing, 2013,22(9): 3366-3378.
[4]	MERKLE P , MLLER K , ZHAO X ,et al. CE6.H results on simplified wedgelet search for DMM Modes 1 and 3[C]// JCT3V Meeting,JCT3V-B0039. 2012: 1-4.
[5]	LEE J Y , PARK M W , KIM C . 3D-CE1:depth intra skip (DIS) mode[C]// JCT3V meeting,JCT3V-K0033. 2005: 1-5.
[6]	LIU H , CHEN Y . Generic segment-wise DC for 3D-HEVC depth intra coding[C]// 2014 IEEE International Conference on Image Processing (ICIP). IEEE, 2014: 3219-3222.
[7]	SHEN L , LI K , FENG G ,et al. Efficient intra mode selection for depth-map coding utilizing spatiotemporal,inter-component and inter-view correlations in 3D-HEVC[J]. IEEE Transactions on Image Processing, 2018,27(9): 4195-4206.
[8]	SALDANHA M , SANCHEZ G , MARCON C ,et al. Fast 3D-HEVC depth map encoding using machine learning[J]. IEEE Transactions on Circuits and Systems for Video Technology,doi:10.1109/TCSVT.2019.2898122, 2019.
[9]	GWEON R H , LEE Y L . Early termination of CU encoding to reduce HEVC complexity[C]// JCTVC Meeting,JCTVC-F045. 2011: 1-4.
[10]	SHEN L , AN P , ZHANG Z ,et al. A 3D-HEVC fast mode decision algorithm for real-time applications[J]. ACM Transactions on Multimedia Computing Communications and Applications, 2015,11(3): 1-23.
[11]	ZHANG Q , HUANG K , WANG X ,et al. Efficient multiview video plus depth coding for 3D-HEVC based on complexity classification of the treeblock[J]. Journal of Real-Time Image Processing,doi:10.1007/S11554-017-0692-5, 2017.
[12]	TOHIDYPOUR H R , POURAZAD M T , NASIOPOULOS P . Online-learning based complexity reduction scheme for 3D-HEVC[J]. IEEE Transactions on Circuits and Systems for Video Technology, 2016,26(10): 1870-1883.
[13]	ZHANG Q , WEI S , SU R . Low-complexity texture video coding based on motion homogeneity for 3D-HEVC[J]. Scientific Programming, 2019: 1-13.
[14]	LI Y , YANG G , ZHU Y ,et al. Hybrid stopping model-based fast PU and CU decision for 3D-HEVC texture coding[J]. Journal of Re-al-Time Image Processing,doi:10.1007/ s11554-019-00876-9, 2019.
[15]	CHEN J , WANG B , LIAO J ,et al. Fast 3D-HEVC inter mode decision algorithm based on the texture correlation of viewpoints[J]. Multime-dia Tools and Applications,doi:10.1007/ s11042-018-6832-5, 2018.
[16]	LEI J , DUAN J , WU F ,et al. Fast mode decision based on grayscale similarity and inter-view correlation for depth map coding in 3D-HEVC[J]. IEEE Transactions on Circuits and Systems for Video Technology, 2018,28(3): 706-718.
[17]	LIAO Y , CHEN M , YEH C ,et al. Efficient inter-prediction depth coding algorithm based on depth map segmentation for 3D-HEVC[J]. Multimedia Tools and Applications, 2019,78(8): 10181-10205.
[18]	PAN Z , LEI J , ZHANG Y ,et al. Adaptive fractional-pixel motion estimation skipped algorithm for efficient HEVC motion estimation[J]. ACM Transactions on Multimedia Computing Communications and Applications, 2018,14(1): 1-19.
[19]	YANG J , KIM J , WON K ,et al. Early SKIP detection for HEVC[C]// JCTVC Meeting,JCTVC-G543. 2011: 1-6.
[20]	LI Y , YANG G , ZHU Y ,et al. Unimodal stopping model based early SKIP mode decision for high efficiency video coding[J]. IEEE Transactions Multimedia, 2017,19(7): 1431-1441.
[21]	PAN Z , KWONG S , SUN M T ,et al. Early merge mode decision based on motion estimation and hierarchical depth correlation for HEVC[J]. IEEE Transactions on Broadcasting, 2014,60(2): 405-412.
[22]	ZHANG N , ZHAO D , CHEN Y ,et al. Fast encoder decision for texture coding in 3D-HEVC[J]. Signal Processing:Image Communication, 2014,29(9): 951-961.
[23]	ZHANG Q , WU Q , WANG X ,et al. Early SKIP mode decision for three dimensional high efficiency video coding using spatial and interview correlations[J]. Journal of Electronic Imaging, 2014,23(5): 1-8.
[24]	SONG Y , JIA K . Early merge mode decision for texture coding in 3D-HEVC[J]. Journal of Visual Communication and Image Representation, 2015,23: 60-68.
[25]	CHEN H , FU C H , ZHANG Y ,et al. Early merge mode decision for depth maps in 3D-HEVC[C]// 22nd International Conference on Digital Signal Processing (DSP). 2017: 1-5.
[26]	LI Y , YANG G , ZHU Y ,et al. Probability model-based early merge mode decision for dependent views coding in 3D-HEVC[J]. ACM Transactions on Multimedia Computing Communications and Applications, 2018,14(4): 1-15.
[27]	MULLER K , VETRO A . Common test conditions of 3DV core experiments[C]// JCT3V Meeting,JCT3V-G1100. 2014: 1-7.
[28]	BJONTEGAARD G . Calculation of average PSNR differences between RD curves[R]. VCEG-M33,Austin:Video Coding Experts Group, 2001.

视频序列	分辨率	帧率/(frame.s^-1)	独立视点＆非独立视点
Balloons	1 024×768	30	V₃＆V₁,V₅
Kendo	1 024×768	30	V₃＆V₁,V₅
Newspaper	1 024×768	30	V₄＆V₂,V₆
GT_Fly	1 920×1 088	25	V₅＆V₁,V₉
Poznan_Hall2	1 920×1 088	25	V₆＆ V₅,V₇
Poznan_Street	1 920×1 088	25	V₄＆ V₃,V₅
Undo_Dancer	1 920×1 088	25	V₅＆V₁,V₉
Shark	1 920×1 088	30	V₅＆V₁,V₉

视频序列	video₀	video₁	video₂	V/V	V/T	S/T	TS_texture	TS_depth	TS_total
Balloons	0.5%	0.8%	0.8%	0.6%	0.6%	0.7%	39.3%	21.3%	32.1%
Kendo	0.6%	0.9%	0.5%	0.7%	0.6%	0.7%	37.1%	18.2%	29.2%
Newspaper	0.5%	0.5%	0.3%	0.5%	0.3%	0.7%	44.1%	21.3%	33.9%
GT_Fly	1.8%	1.9%	2.1%	1.9%	1.8%	1.8%	48.7%	26.9%	39.7%
Poznan_Hall2	0.6%	0.8%	0.9%	1.0%	0.9%	0.8%	50.1%	23.1%	38.6%
Poznan_Street	0.2%	0.7%	0.3%	0.6%	0.7%	0.4%	47.5%	26.5%	38.1%
Undo_Dancer	1.0%	0.7%	0.6%	1.1%	0.9%	1.2%	37.3%	30.6%	34.6%
Shark	0.2%	-0.1%	-0.1%	0.2%	-0.1%	0.4%	31.2%	26.4%	28.9%
平均值	0.7%	0.8%	0.7%	0.8%	0.7%	0.8%	41.9%	24.3%	34.4%

视频序列	video₀	video₁	video₂	V/V	V/T	S/T	TS_texture	TS_depth	TS_total
alloons	0.1%	1.1%	1.2%	0.6%	0.5%	0.6%	20.8%	11.3%	17.0%
Kendo	0.2%	1.9%	1.8%	0.9%	0.8%	0.7%	19.6%	8.9%	15.1%
Newspaper	0.1%	2.9%	0.8%	0.9%	0.6%	1.0%	30.0%	10.2%	21.2%
GT_Fly	0.7%	-1.0%	-0.8%	0.8%	0.6%	0.4%	20.6%	12.0%	23.0%
Poznan_Hall2	0.1%	1.4%	3.1%	0.9%	0.8%	0.8%	26.9%	11.4%	20.2%
Poznan_Street	0.1%	0.4%	0.5%	0.4%	0.3%	0.4%	33.0%	13.2%	24.2%
Undo_Dancer	0.4%	-0.8%	-0.9%	0.5%	0.3%	0.2%	26.2%	16.8%	22.5%
Shark	0.3%	-1.7%	-1.5%	0.2%	0.0%	0.2%	28.8%	11.9%	21.5%
平均值	0.3%	0.5%	0.5%	0.7%	0.5%	0.5%	25.7%	12.0%	20.6%

视频序列	video₀	video₁	video₂	V/V	V/T	S/T	TS_texture	TS_depth	TS_total
Balloons	0.5%	1.0%	0.9%	0.7%	0.7%	0.5%	36.3%	6.0%	24.5%
Kendo	0.5%	1.2%	0.9%	0.8%	0.8%	0.5%	35.2%	3.7%	22.4%
Newspaper	0.3%	0.4%	0.6%	0.4%	0.4%	0.3%	37.9%	5.1%	23.6%
GT_Fly	1.5%	2.6%	2.8%	2.0%	1.9%	1.6%	37.0%	6.1%	24.6%
Poznan_Hall2	0.6%	0.8%	0.6%	0.7%	0.7%	0.6%	33.1%	4.1%	20.9%
Poznan_Street	0.2%	0.9%	1.3%	0.6%	0.6%	0.4%	34.2%	5.7%	21.8%
Undo_Dancer	0.7%	4.1%	4.7%	2.0%	1.9%	1.3%	32.4%	9.1%	23.5%
Shark	0.7%	1.6%	1.8%	1.1%	1.0%	0.8%	31.2%	5.6%	20.3%
平均值	0.6%	1.6%	1.7%	1.0%	1.0%	0.8%	34.6%	5.7%	22.7%

视频序列	本文提出的方法		文献[16]方法		文献[17]方法
视频序列	S/T	TS_Depth	S/T	TS_Depth	S/T	TS_depth
Balloons	0.1%	33.6%	2.4%	41.5%	0.8%	16.3%
Kendo	0.1%	28.4%	4.0%	35.5%	0.3%	16.2%
Newspaper	0.3%	32.4%	2.0%	45.0%	0.7%	15.2%
GT_Fly	0.1%	39.4%	1.6%	42.1%	—	—
Poznan_Hall2	0.2%	40.0%	0.9%	44.8%	-0.6%	17.3%
Poznan_Street	0.1%	40.0%	2.1%	47.2%	0.9%	17.0%
Undo_Dancer	0.4%	44.8%	2.6%	48.5%	—	—
Shark	0.2%	39.0%	3.0%	31.6%	0%	16.5%
平均值	0.2%	37.2%	2.3%	42.0%	0.4%	16.4%

基于学习模型的3D-HEVC提前Merge模式终止算法

Early Merge mode decision algorithm for 3D-HEVC based on learning model

在线阅读

PDF下载

可视化

摘要/Abstract

引用本文

使用本文

图/表 11

参考文献 28

相关文章 1

Metrics

推荐阅读 0