基于纹理平滑度的视点合成失真优化快速算法

doi:10.11959/j.issn.1000-436x.2016057

摘要/Abstract

摘要：

针对3D-HEVC 中深度图编码采用的视点合成失真优化方法的高复杂度问题，提出一种基于纹理平滑度的快速算法。首先结合帧内DC预测特性和统计学方法分析平坦纹理图中像素规律并设定基于纹理图平坦度的跳过准则；然后在深度图编码采用视点合成失真优化方法时提前分离出纹理图平坦区域所对应的深度图区域，并终止该区域像素基于虚拟视点合成的视点合成失真计算过程。实验结果证明该算法的有效性，能在保持编码质量的同时减少大量编码时间。

关键词: 3D-HEVC, 深度图编码, 视点合成失真优化, 纹理图平坦度

Abstract:

A fast algorithm was proposed in order to reduce the v ew synthesis optimization (VSO) process in depth cod-ing for 3D-HEVC based on texture map smoothness.With the coding nformation derived from texture video sequences,the pixel regularity of smooth texture region was analyzed to set the skip rule using the properties of intra DC prediction and statistical methods.Then the depth regions corres to the flat texture map regions could be extracted and the VSO process of pixels belonging to this type of depth could be skipped.Experimental results show the effective-ness of the proposed algorithm.

Key words: 3D-HEVC, depth coding, view synthesis optimization, texture smoothness

窦环,贾克斌,陈锐霖,萧允治,吴强. 基于纹理平滑度的视点合成失真优化快速算法[J]. 通信学报, 2016, 37(3): 98-106.

Huan DOU,Ke-bin JIA,Rui-lin CHEN,Yun-zhi XIAO,Qiang WU. Fast view synthesis optimization algorithm based on texture smoothness[J]. Journal on Communications, 2016, 37(3): 98-106.

图/表 9

图1

图2

图3

表1

表2

表3

表4

不同算法的深度图编码时间比较结果"

序列	QP	不同方法的编码时间/s				各方法占ori比例/%			prop占文献[9] /%比例	prop ES占比例/%
序列	QP	ori	ES	文献[9]	prop	文献[9]	ES	prop	prop占文献[9] /%比例	prop ES占比例/%
Balloons	25	8 769.607	5 654.49	7 201.647	5 175.079	82.12	64.48	59.01	71.86	91.52
	30	7 046.539	4 521.449	5 812.683	4 133.457	82.49	64.17	58.66	71.11	91.42
	35	6 373.172	4 157.844	5 308.588	3 729.521	83.30	65.24	58.52	70.25	89.70
	40	5 897.246	3 991.447	5 030.381	3 441.448	85.30	67.68	58.36	68.41	86.22
Kendo	25	10 428.268	6 830.559	8 648.402	5 674.497	82.93	65.50	54.41	65.61	83.08
	30	8 257.474	5 447.459	6 846.684	4 430.036	82.91	65.97	53.65	64.70	81.32
	35	7 315.150	4 875.937	6 094.086	3 881.701	83.31	66.66	53.06	63.70	79.61
	40	6 543.406	4 470.179	5 760.688	3 565.596	88.04	68.32	54.49	61.90	79.76
Newspaper	25	7 939.096	5 621.668	6 810.838	5 314.792	85.79	70.81	66.94	78.03	94.54
	30	6 658.800	4 737.368	5 739.726	4 374.851	86.20	71.14	65.70	76.22	92.35
	35	6 225.230	4 435.27	5 399.458	4 017.670	86.74	71.25	64.54	74.41	90.58
	40	5 878.834	4 233.907	5 145.484	3 716.928	87.53	72.02	63.23	72.24	87.79
GTFly	25	20 825.208	12 566.592	17 100.368	11 084.931	82.11	60.34	53.23	64.82	88.21
	30	15 750.355	9 938.751	13 619.904	8 387.293	86.47	63.10	53.25	61.58	84.39
	35	13 889.304	9 419.958	12 676.683	7 272.111	91.27	67.82	52.36	57.37	77.20
	40	12 394.614	8 524.951	11 556.152	6 321.666	93.24	68.78	51.00	54.70	74.15
PoznanStreet	25	14 286.545	9 717.326	11 584.369	8 813.954	81.09	68.02	61.69	76.08	90.70
	30	11 914.260	8 186.395	9 889.045	7 058.361	83.00	68.71	59.24	71.38	86.22
	35	11 478.971	8 087.928	9 972.184	6 571.761	86.87	70.46	57.25	65.90	81.25
	40	10 604.318	7 552.166	9 427.314	5 846.729	88.90	71.22	55.14	62.02	77.42
UndoDancer	25	17 372.154	10 704.598	13 164.029	9 940.111	75.78	61.62	57.22	75.51	92.86
	30	14 078.756	8 993.902	11 300.668	8 111.834	80.27	63.88	57.62	71.78	90.19
	35	12 285.994	8 205.154	10 390.345	7 151.651	84.57	66.78	58.21	68.83	87.16
	40	11 816.451	8 027.814	10 179.028	6 394.110	86.14	67.94	54.11	62.82	79.65
		平均值				84.85	67.16	57.54	67.97	85.72

表4

表5

表6

参考文献 20

[1]	MULLER K , SCHWARZ H , MARPE D , et al. 3D high-efficiency video coding for multi-view video and depth data[J]. IEEE Transac-tions on Image Processing, 2013,22(9): 3366-3378.
[2]	ISO/IEC JTC 1/SC 29/WG 11 (MPEG).High efficiency video Coding[S].ITU-T Rec H.265 and ISO/IEC 23008-2 (HEVC),ITU-T and ISO/IEC, 2013.
[3]	SULLIVAN G J , OHM J , HAN W J , et al. Overview of the h gh effi-ciency video coding (HEVC) standard[J]. IEEE Transacti s on Cir-cuits and Systems for Video Technology, 2012,22(12): 1649-1668.
[4]	CHEN Y , VETRO A . Next-generation 3D formats with depth map support[J]. MultiMedia IEEE, 2014,21(2): 90-94.
[5]	SMOLIC A , MULLER K , DIX K , et al. Intermediate view in la-tion based on multiview video plus depth for advanced 3D video sys-tems[C]// IEEE International Conference on Image Processing. San Diego,CA, c2008:2448-2451.
[6]	MERKLE P , MORVAN Y , SMOLIC A , et al. The effect of depth com-pression on multiview rendering quality[C]// 3DTV Conference:The True Vision-Capture,Transmission and Display of 3D Video. Istanbul, c2008:245-248.
[7]	TECH G , SCHWARZ H , MULLER K , et al. 3D video coding us ng the synthesized view distortion change[C]// Picture Cod ng Symposium(PCS). Krakow, c2012:25-28.
[8]	OH B , LEE J , PARK D , et al. 3D-CE8.h results on view synthesis optimization by Samsung,HHI and LG-PKU[R]. ITU-T SG 16 WP 3 and ISO/IEC JTC 1/SC 29/WG 11,1st Meeting,JCT3V-A0093, Stockholm,Sweden, 2012.
[9]	MA S , WANG S , GAO W . Low complexity adaptive view synthesis optimization in HEVC based 3D video coding[J]. IEEE Transactions on Multimedia, 2014,16(1): 266-271.
[10]	MA S , WANG S , GAO W . Zero-synthesis view difference aware view synthesis optimization for HEVC based 3D video co pres-sion[J]. Visual Communications ＆ Image Processing,IEEE, 2012,333(6): 1-6.
[11]	WANG S , MA S , LIU H , et al. CE8.h:results of simplification of view synthesis optimization by detection of zero distortion change in syn-thesized view[R]// 1st JCT3V Meeting,ITU-T SG 16 WP 3 and ISO/IEC JTC 1/SC 29/WG 11 JCT2-A0083, Stockholm,SE, 2012.
[12]	OH B T , LEE J , PARK D S . Depth map coding based on synthesized view distortion function[J]. IEEE Journal of Selected opics in Signal Processing, 2011,5(7): 1344-1352.
[13]	OH B T , OH K J . View synthesis distortion estimation for avc-and hevc-compatible 3D video coding[J]. IEEE Transactions on Circuits and Systems for Video Technology, 2014,24(6): 1006-1015.
[14]	ZHANG Q , TIAN L , HUANG L , et al. Rendering distortion estimation model for 3D high efficiency depth coding[J]. Mathematical Problems in Engineering, 2014(1): 1-7.
[15]	LI C , JIN X , DAI Q . A novel distortion model for depth coding in 3D-HEVC[C]// IEEE International Conference on Image Processing. Paris, c2014:3228-3232.
[16]	LI C , JIN X , DAI Q . An efficient distortion model configuration for depth coding in 3D-HEVC[C]// IEEE Asia Pacific Conference on Cir-cuits and Systems. Ishigaki, c2014:9-12.
[17]	TECH G , WEGNER K , CHEN Y , et al. Test Model 8 of 3D-HEVC and MV-HEVC[R]// 8th JCT3V Meeting,ITU-T SG 16 WP 3 and ISO/IEC JTC 1/SC 29/WG 11,JCT3V-H1003, Valencia,ES, 2014.
[18]	FRANK B , DAVID F , KARSTEN S . HEVC 3D extension software[EB/OL]. .
[19]	MULLER K , VETRO A . Common test conditions of 3DV core expe-riments[R]. JCT3V meeting,JCT3VG1100, San Jose,US, 2014.
[20]	BJONTEGAARDA G . Calculation of average psnr differences be-tween rd-curves[R]. ITU-Telecommunications Standardization Secto vceg-m33, 2001.

测试序列	视点序号	命中率/%
Newspaper	1	14.3
	3	77.2
	5	99.1
Balloons	2	60.4
	4	33.2
	6	63.6
Kendo	1	7.4
	3	12.1
	5	22.2
GTFly	1	100
	5	19.9
	9	90.9
UndoDancer	1	27.0
	5	69.2
	9	65.8
PoznanStreet	3	92.9
	4	92.1
	5	97.0

序列		平滑像素行百分比/%
序列	QP=25	QP=30	QP=35	QP=40
Newspaper	38.81	47.32	54.25	62.17
Balloons	50.03	57.73	64.53	70.49
Kendo	32.73	43.00	50.25	55.47
GTFly	40.06	52.55	67.72	74.57
UndoDancer	37.51	50.70	61.86	66.79
PoznanStreet	33.67	44.19	53.57	63.86

序列	QP	文献[9]	ES	prop
Balloons	25	64.31%	63.63%	77.70%
	30	62.28%	61.65%	79.73%
	35	58.03%	57.46%	80.47%
	40	49.55%	49.01%	80.66%
Kendo	25	62.78%	61.95%	80.97%
	30	61.10%	60.33%	93.11%
	35	56.70%	55.91%	84.27%
	40	49.97%	49.14%	84.90%
Newspaper	25	50.17%	49.63%	65.78%
	30	46.49%	46.01%	68.80%
	35	41.56%	41.00%	70.27%
	40	37.42%	36.74%	71.60%
UndoDancer	25	69.13%	67.37%	78.13%
	30	63.11%	61.36%	77.90%
	35	55.48%	53.37%	77.92%
	40	45.28%	43.45%	79.70%
GTFly	25	67.35%	66.10%	79.66%
	30	62.97%	61.30%	81.54%
	35	52.97%	50.40%	83.99%
	40	48.84%	46.08%	86.29%
PoznanStreet	25	53.13%	52.44%	70.21%
	30	48.74%	48.19%	74.48%
	35	44.59%	43.04%	78.18%
	40	40.71%	39.08%	79.80%
平均值		53.86%	52.69%	78.59%