基于随机森林的HEVC复杂度控制方法

doi:10.11959/j.issn.1000-0801.2019034

电信科学 ›› 2019, Vol. 35 ›› Issue (2): 14-26.doi: 10.11959/j.issn.1000-0801.2019034

基于随机森林的HEVC复杂度控制方法

文鹏,彭宗举,陈芬,蒋刚毅,郁梅

宁波大学信息科学与工程学院，浙江宁波 315211

修回日期:2019-01-20 出版日期:2019-02-01 发布日期:2019-02-23
作者简介:文鹏（1994- ），男，宁波大学信息科学与工程学院硕士生，主要研究方向为三维视频信号编码。|彭宗举（1973- ），男，博士，宁波大学信息科学与工程学院教授，主要研究方向为三维视频信号处理与编码。|陈芬（1973- ），女，博士，宁波大学信息科学与工程学院教授，主要研究方向为三维视频信号处理与编码。|蒋刚毅（1964- ），男，博士，宁波大学信息科学与工程学院教授，主要研究方向为三维视频信号处理与编码。|郁梅（1968- ），女，博士，宁波大学信息科学与工程学院教授，主要研究方向为三维视频信号处理与编码。
基金资助:
国家自然科学基金资助项目(61771269);国家自然科学基金资助项目(61620106012);国家自然科学基金资助项目(61671258);浙江省自然科学基金资助项目(LY17F010005);宁波市自然科学基金资助项目(2018A610052)

Complexity control method of random forest based HEVC

Peng WEN,Zongju PENG,Fen CHEN,Gangyi JIANG,Mei YU

Faculty of Information Science and Engineering,Ningbo University,Ningbo 315211,China

Revised:2019-01-20 Online:2019-02-01 Published:2019-02-23
Supported by:
The National Natural Science Foundation of China(61771269);The National Natural Science Foundation of China(61620106012);The National Natural Science Foundation of China(61671258);The Natural Science Foundation of Zhejiang Province of China(LY17F010005);The Natural Science Foundation of Ningbo of China(2018A610052)

摘要/Abstract

摘要：

高效视频编码（high efficiency video coding，HEVC）具有较高的计算复杂度，而快速编码算法不能在给定编码时间下完成视频编码，因此提出了一种基于随机森林的HEVC复杂度控制方法。首先训练得到3种不同预测准确率的随机森林分类器，实现编码树单元（coding tree unit，CTU）的多种编码配置；然后，通过建立平均深度—复杂度模型为CTU分配复杂度；最后通过平坦度、平均深度、比特和CTU级累计编码误差确定的CTU编码配置完成复杂度控制。实验结果表明，本文算法复杂度控制精度较高，在率失真性能方面优于当前复杂度控制方法。

关键词: 复杂度控制, 高效视频编码, 随机森林

Abstract:

High efficiency video coding (HEVC) has high computational complexity,and fast algorithm cannot perform video coding under restricted coding time.Therefore,a complexity control method of HEVC based on random forest was proposed.Firstly,three random forest classifiers with different prediction accuracy were trained to provide various coding configurations for coding tree unit (CTU).Then,an average depth-complexity model was built to allocate CTU complexity.Finally,the CTU coding configuration,determined by the smoothness,average depth,bit,and CTU-level accumulated coding error,was used to complete complexity control.The experimental results show that the proposed method has better complexity control precision,and outperforms the state-of-the-art method in terms of video quality.

Key words: complexity control, HEVC, random forest

中图分类号:

TN919.8

文鹏,彭宗举,陈芬,蒋刚毅,郁梅. 基于随机森林的HEVC复杂度控制方法[J]. 电信科学, 2019, 35(2): 14-26.

Peng WEN,Zongju PENG,Fen CHEN,Gangyi JIANG,Mei YU. Complexity control method of random forest based HEVC[J]. Telecommunications Science, 2019, 35(2): 14-26.

图/表 12

图1

图2

图3

表1

表2

图4

图5

表3

FS> 2 3 时CTU的编码配置"

编码配置	e_n <-MT	$- MT \leq e_{n} < - \frac{MT}{2}$	$- \frac{MT}{2} \leq e_{n} \leq \frac{MT}{2}$	$- \frac{MT}{2} < e_{n} \leq MT$	e_n>MT
Bit≥10?MB	ET12	正常编码	正常编码	正常编码	正常编码
	ET22
MB≤Bit<10?MB	ET02	ET02	正常编码	正常编码	正常编码
	skip12,ET12	ET12
	skip22,ET22	ET22
0.1?MB≤Bit<MB	ET02	ET02	ET02	ET02	正常编码
	skip12,ET11	ET11	ET12	ET12
	skip22,ET21	ET21	ET21	ET22
Bit≤0.1?MB	Only0	ET00	ET02	ET02	正常编码
		ET10	ET10	ET12
		ET20	ET20	ET20

表3

表4

FS≤ 2 3 时CTU的编码配置"

编码配置	e_n <-MT	$- MT \leq e_{n} < - \frac{MT}{2}$	$- \frac{MT}{2} \leq e_{n} \leq \frac{MT}{2}$	$- \frac{MT}{2} < e_{n} \leq MT$	e_n>MT
MD≤0.5	Only0	skip02,ET01	skip2,ET02	skip02,ET02	正常编码
		skip12,ET10	skip2,ET10	skip12,ET11
		skip22,ET20	skip2,ET20	skip22,ET20
0.5<MD≤1.5	Only12	skip00,ET02	skip00,ET02	skip01,ET02	正常编码
	skip11	skip11,ET11	skip12,ET12	skip12,ET12
		skip22,ET20	skip22,ET20	skip22,ET22
1.5<MD≤2.5	Only123	skip00,ET02	skip00,ET02	skip00,ET02	正常编码
	skip10,ET11	skip11,ET11	skip12,ET11	skip12,ET12
	skip21,ET20	skip21,ET20	skip22,ET21	skip22,ET22
MD>2.5	Only23	skip00,ET02	skip00,ET02	skip00,ET02	正常编码
	skip21,ET20	skip10,ET12	skip11,ET12	skip11,ET12
		skip22,ET20	skip22,ET21	skip22,ET22

表4

表5

表6

表7

参考文献 22

[1]	SULLIVAN G J , OHM J R , HAN W J ,et al. Overview of the high efficiency video coding (HEVC) standard[J]. IEEE Transactions on Circuits ＆ Systems for Video Technology, 2012,22(12): 1649-1668.
[2]	WIEGAND T , SULLIVAN G J , BJ?NTEGAARD G ,et al. Overview of the H.264/AVC video coding standard[J]. IEEE Transactions on Circuits ＆ Systems for Video Technology, 2003,13(7): 560-576.
[3]	SHAN Y , YANG E H . Fast HEVC intra coding algorithm based on machine learning and Laplacian transparent composite model[C]// IEEE International Conference on Acoustics,Speech and Signal Processing,March 5-9,2017,New Orleans,LA,USA. Piscataway:IEEE Press, 2017: 2642-2646.
[4]	ZHU L , ZHANG Y , PAN Z ,et al. Binary and multi-class learning based low complexity optimization for HEVC encoding[J]. IEEE Transactions on Broadcasting, 2017,63(3): 547-561.
[5]	TAN T K , WEERAKKODY R , MRAK M ,et al. Video quality evaluation methodology and verification testing of HEVC compression performance[J]. IEEE Transactions on Circuits ＆ Systems for Video Technology, 2016,26(1): 76-90.
[6]	ZHANG T , SUN M T , ZHAO D ,et al. Fast intra mode and CU size decision for HEVC[J]. IEEE Transactions on Circuits ＆Systems for Video Technology, 2017,27(8): 1714-1726.
[7]	MALLIKARACHCHI T , TALAGALA D S , ARACHCHI H K ,et al. Content-adaptive feature-based CU size prediction for fast low-delay video encoding in HEVC[J]. IEEE Transactions on Circuits ＆ Systems for Video Technology, 2018,28(3): 693-705.
[8]	TSAI T H , SU S S , LEE T Y . Fast mode decision method based on edge feature for HEVC inter prediction[J]. IET Image Processing, 2018,12(5): 644-651.
[9]	姚英彪, 李晓娟 . 基于图像空间相关性与纹理的HEVC块划分快速算法[J]. 电信科学, 2015,31(1): 32-40.
	YAO Y B , LI X J . Fast algorithm for HEVC block partitioning based on image spatial correlation and texture[J]. Telecommu-nications Science, 2015,31(1): 32-40.
[10]	CORREA G , ASSUNCAO P , CRUZ L A D S ,et al. Computational complexity control for HEVC based on coding tree spatio-temporal correlation[C]// IEEE 20th International Conference on Electronics,Circuits,and Systems,Dec 8-11,2014,Abu Dhabi,United Arab Emirates. Piscataway:IEEE Press, 2014: 937-940.
[11]	CORREA G , ASSUNCAO P , AGOSTINI L ,et al. Coding tree depth estimation for complexity reduction of HEVC[C]// Data Compression Conference,March 20-22,2013,Snowbird,UT,USA. Piscataway:IEEE Press, 2013: 43-52.
[12]	CORREA G , ASSUNCAO P , AGOSTINI L ,et al. Complexity scalability for real-time HEVC encoders[J]. Journal of Real-Time Image Processing, 2016,12(1): 107-122.
[13]	CORREA G , ASSUNCAO P , CRUZ L A D S ,et al. Encoding time control system for HEVC based on rate-distortion-complexity analysis[C]// IEEE International Symposium on Circuits and Systems,May 24-27,2015,Lisbon,Portugal. Piscataway:IEEE Press, 2015: 1114-1117.
[14]	JIMéNEZ-MORENO A ， MARTíNEZ-ENRíQUEZ E , DíAZ-DE-MARíA F . Complexity control based on a fast coding unit decision method in the HEVC video coding standard[J]. IEEE Transactions on Multimedia, 2016,18(4): 563-575.
[15]	ZHANG J , KWONG S , ZHAO T ,et al. CTU-level complexity control for high efficiency video coding[J]. IEEE Transactions on Multimedia, 2018,20(1): 29-44.
[16]	DENG X , XU M , JIANG L ,et al. Subjective-driven complexity control approach for HEVC[J]. IEEE Transactions on Circuits＆ Systems for Video Technology, 2016,26(1): 91-106.
[17]	DENG X , XU M , LI C . Hierarchical complexity control of HEVC for live video encoding[J]. IEEE Access, 2016,4(99): 7014-7027.
[18]	徐博文, 王晓东, 郭磊 ,等. 基于幻方变换的 HEVC 熵编码加密方案[J]. 电信科学, 2018,34(3): 95-104.
	XU B W , WANG X D , GUO L ,et al. A video encryption in HEVC entropy coding based on magic square[J]. Telecommu-nications Science, 2018,34(3): 95-104.
[19]	BREIMAN L . Random forests[J]. Machine Learning, 2001,45(1): 5-32.
[20]	BREIMAN L . Bagging predictors[J]. Machine Learning, 1996,24(2): 123-140.
[21]	BREIMAN L I , FRIEDMAN J H , OLSHEN R A ,et al. Classification and regression trees (CART)[J]. Encyclopedia of Ecology, 1984,40(3): 582-588.
[22]	BOSSEN F , . Common test conditions and software reference configurations[C]// Document JCTVC-F900,Joint Collaborative Team on Video Coding (JCT-VC) of ITU-T SG16 WP3 and ISO/IEC JTC1/SC29/WG11 6th Meeting,July 14-22,2011,Torino,Italy.[S.l.:s.n]. 2011.

深度	skip		ET
深度	W₀/W₁	P₀	W₀/W₁	P₀
0	1:10	66.4%	1:10	84.9%
0	2:1	93.1%	1:1	90.7%
0	10:1	97.4%	10:1	98.9%
1	1:10	56.5%	1:10	78.2%
1	3:1	92.3%	1:1	90.3%
1	10:1	96.7%	10:1	98.6%
2	1:10	60.6%	1:10	72.7%
2	3:1	90.9%	1:1	91.5%
2	10:1	97.1%	10:1	99.1%

	P(s\|MD)	s₀	s₁	s₂	s₃
BQMall	MD≤0.5	100	0	0	0
	0.5<MD≤1.5	0	80.35	16.78	2.87
	1.5<MD≤2.5	0	28.13	48.40	23.48
	MD>2.5	0	0	35.52	64.48
Johnny	MD≤0.5	100	0	0	0
	0.5<MD≤1.5	0	85.14	13.81	1.05
	1.5<MD≤2.5	0	38.12	49.37	12.51
	MD>2.5	0	0	37.74	62.26
ParkScene	MD≤0.5	100	0	0	0
	0.5<MD≤1.5	0	81.58	16.51	1.91
	1.5<MD≤2.5	0	27.54	51.78	20.67
	MD>2.5	0	0	35.03	64.97

序列（类）	T_c =80%			T_c =60%			T_c =40%
序列（类）	R_c	?PSNR/dB	?BR	R_c	?PSNR/dB	?BR	R_c	?PSNR/dB	?BR
BasketballDrive(B)	81.01%	-0.01	0.09%	61.71%	-0.04%	0.25	38.77%	-0.10	2.82%
BQTerrace (B)	80.09%	-0.01	-0.56%	60.37%	-0.03%	-0.94	40.95%	-0.06	-0.33%
ParkScene (B)	81.45%	-0.01	-0.02%	62.17%	-0.04%	0.07	40.66%	-0.08	1.16%
Johnny (E)	79.97%	-0.01	-0.50%	60.54%	-0.02%	-0.63	40.65%	-0.05	0.25%
FourPeople (E)	79.46%	-0.02	-0.40%	58.66%	-0.03%	-0.62	39.57%	-0.07	1.45%
Vidyo1 (E)	79.67%	-0.01	-0.23%	59.56%	-0.02%	-0.40	40.33%	-0.06	0.71%
Vidyo3 (E)	80.59%	-0.01	0.09%	60.72%	-0.02%	0.19	41.64%	-0.07	1.66%
Vidyo4 (E)	80.08%	-0.01	0.04%	60.36%	-0.02%	0.07	41.69%	-0.06	1.32%
PartyScene (C)	82.20%	-0.01	0.42%	61.31%	-0.04%	0.83	40.72%	-0.08	3.79%
RaceHorsesC (C)	82.62%	0.00	0.31%	62.16%	-0.03%	0.71	41.54%	-0.09	4.51%
BasketballDrill (C)	81.62%	-0.01	0.18%	60.77%	-0.03%	0.54	41.27%	-0.10	4.15%
SlideEditing (F)	78.69%	-0.02	0.20%	60.38%	-0.02%	-0.06	40.20%	-0.04	0.32%
SlideShow (F)	80.05%	-0.03	-0.04%	60.81%	-0.05%	0.33	40.90%	-0.22	2.65%
平均值	80.58%	-0.01	-0.03%	60.73%	-0.03%	0.03	40.68%	-0.08	1.88%

序列（类）	T_c =80%						T_c =60%
	本文算法			参考文献[16]算法			本文算法			参考文献[16]算法
	R_c	?PSNR/dB	?BR	R_c	?PSNR/dB	?BR	R_c	?PSNR/dB	?BR	R_c	?PSNR/dB	?BR
BasketballDrive (B)	81.01%	-0.01	0.09%	78.58%	-0.03	0.33%	61.71%	-0.04	0.25%	60.24%	-0.15	2.02%
BQTerrace (B)	80.09%	-0.01	-0.56%	79.75%	-0.02	0.33%	60.37%	-0.03	-0.94%	61.90%	-0.08	3.20%
ParkScene (B)	81.45%	-0.01	-0.02%	80.00%	-0.03	0.60%	62.17%	-0.04	0.07%	61.27%	-0.14	5.87%
Johnny (E)	79.97%	-0.01	-0.50%	80.84%	0.00	-0.05%	60.54%	-0.02	-0.63%	59.10%	0.00	-0.02%
FourPeople (E)	79.46%	-0.02	-0.40%	80.93%	0.00	0.13%	58.66%	-0.03	-0.62%	60.23%	-0.01	0.44%
Vidyo1 (E)	79.67%	-0.01	-0.23%	81.98%	0.00	0.03%	59.56%	-0.02	-0.40%	62.19%	-0.02	0.62%
Vidyo3 (E)	80.59%	-0.01	0.09%	81.11%	0.00	0.04%	60.72%	-0.02	0.19%	63.36%	-0.05	1.55%
Vidyo4 (E)	80.08%	-0.01	0.04%	82.28%	0.00	0.15%	60.36%	-0.02	0.07%	61.01%	-0.01	0.21%
PartyScene (C)	82.20%	-0.01	0.42%	78.73%	-0.10	2.53%	61.31%	-0.04	0.83%	60.50%	-0.25	8.29%
RaceHorsesC (C)	82.62%	0.00	0.31%	78.31%	-0.08	1.56%	62.16%	-0.03	0.71%	58.13%	-0.45	7.37%
BasketballDrill (C)	81.62%	-0.01	0.18%	82.75%	-0.02	0.46%	60.77%	-0.03	0.54%	62.71%	-0.18	4.85%
SlideEditing (F)	78.69%	-0.02	0.20%	81.53%	-0.03	0.42%	60.38%	-0.02	-0.06%	61.06%	-0.30	3.32%
SlideShow (F)	80.05%	-0.03	-0.04%	80.18%	-0.03	0.40%	60.81%	-0.05	0.33%	62.00%	-0.06	0.58%
平均值	80.58%	-0.01	-0.03%	80.54%	-0.02	0.53%	60.73%	-0.03	0.03%	61.05%	-0.13	2.95%

序列（类）	本文算法			参考文献[16]算法
序列（类）	R_c	?PSNR/dB	?BR	R_c	?PSNR/dB	?BR
BasketballDrive (B)	38.77%	-0.10	2.82%	40.37%	-0.28	6.03%
BQTerrace (B)	40.95%	-0.06	-0.33%	41.62%	-0.31	6.59%
ParkScene (B)	40.66%	-0.08	1.16%	40.02%	-0.34	8.24%
Johnny (E)	40.65%	-0.05	0.25%	41.74%	-0.06	2.21%
FourPeople (E)	39.57%	-0.07	1.45%	41.10%	-0.24	6.50%
Vidyo1 (E)	40.33%	-0.06	0.71%	41.20%	-0.05	1.89%
Vidyo3 (E)	41.64%	-0.07	1.66%	38.37%	-0.11	3.67%
Vidyo4 (E)	41.69%	-0.06	1.32%	40.02%	-0.10	3.77%
PartyScene (C)	40.72%	-0.08	3.79%	41.42%	-1.03	13.24%
RaceHorsesC (C)	41.54%	-0.09	4.51%	37.55%	-1.34	11.12%
BasketballDrill (C)	41.27%	-0.10	4.15%	42.72%	-0.27	6.36%
SlideEditing (F)	40.20%	-0.04	0.32%	41.65%	-0.87	6.54%
SlideShow (F)	40.90%	-0.22	2.65%	41.12%	-0.12	1.83%
平均值	40.68%	-0.08	1.88%	40.68%	-0.39	6.00%

基于随机森林的HEVC复杂度控制方法

Complexity control method of random forest based HEVC

在线阅读

PDF下载

可视化

摘要/Abstract

引用本文

使用本文

图/表 12

参考文献 22

相关文章 10

Metrics

推荐阅读 0

[1]	徐嘉, 简志华, 金宏辉, 吴超, 游林, 吴迎笑. 基于中心对称局部二值模式的合成伪装语音检测方法[J]. 电信科学, 2023, 39(1): 72-78.
[2]	于佳祺, 简志华, 徐嘉, 游林, 汪云路, 吴超. 基于联合特征与随机森林的伪装语音检测[J]. 电信科学, 2022, 38(6): 91-99.
[3]	孟翔, 殷海兵, 黄晓峰. 基于统计建模的HEVC快速率失真估计算法[J]. 电信科学, 2021, 37(1): 58-68.
[4]	卢子萌,陈佳怡,李璟,谢岳,蒋欣利,韩蕾,郭倩. 基于加权随机森林算法的空巢电力用户识别方法[J]. 电信科学, 2020, 36(8): 112-121.
[5]	张溶芳,许丹丹,王元光,潘思宇,李正茂. 机器学习在物联网虚假用户识别中的运用[J]. 电信科学, 2019, 35(7): 136-144.
[6]	杜续,冯景瑜,吕少卿,石薇. 基于随机森林回归分析的PM2.5浓度预测模型[J]. 电信科学, 2017, 33(7): 66-75.
[7]	王彦青,王瀚辰. 一种识别骚扰电话的组合算法研究[J]. 电信科学, 2017, 33(7): 112-119.
[8]	李倩,江昊,杨锦涛. 基于手机上网记录数据的个体相遇预测[J]. 电信科学, 2017, 33(10): 115-123.
[9]	刘歌,张国毅,于岩. 基于随机森林的雷达信号脉内调制识别[J]. 电信科学, 2016, 32(5): 69-78.
[10]	王铮,任华,方燕萍. 随机森林在运营商大数据补全中的应用[J]. 电信科学, 2016, 32(12): 7-12.