基于HEVC的3D视频编解码研究

发布时间：2018-11-04 18:27

【摘要】：随着3D电影、3D电视广播、裸眼3D显示技术等3D视频技术的提高,3D视频技术成为了学术界和工业界的研究热点。从2001年起,国际标准化组织ISO的运动图像专家组MPEG发起了制定自由视点电视(Free-viewpoint Television, FTV)国际标准的长期目标,其允许在自由改变观看视点的条件下观看3D视频。其中,多视点视频编码(Multi-view Video Coding, MVC)是FTV制定的第一阶段,三维视频3DV是FTV标准的第二阶段。国际电信联盟ITU的第16工作组VCEG和MPEG共同发起了JCT-3V工作组,专门制定新一代的3D视频编码技术,其中基于新一代高效视频编码技术H.265/HEVC的3D视频编码扩展为3D-HEVC。在3D-HEVC框架中,3D视频以多视点视频加深度信息(Multi-view Video plus Depth, MVD)的形式表示。编码端在H.265/HEVC的基础上对非独立视点和深度信息增加了新的编码工具。本论文对基于HEVC的3D视频编码3D-HEVC方案进行研究。首先,文本详细介绍了基于HEVC的3D视频编码方案3D-HEVC的基本框架和关键技术,并介绍了3D视频编码技术中的三个重要研究难点：纹理图编码、深度图编码和深度估计。其次,本文在3D-HEVC的框架基础上提出了基于深度图的纹理图快速模式判决算法。该算法利用3D视频中深度图与纹理图的关系,根据深度图将相应的纹理图分为前景、主景和背景部分,对相应部分的模式判决进行精简；引入运动矢量与深度值的关系,进行视点间和视点内模式的快速选择；接着,论文着重介绍了3D-HEVC中的深度图编码。根据深度图大范围平滑区域和尖锐边缘的特性,在深度图的帧内预测模式中提出了四种新的预测模式；根据深度图辅助虚拟视点合成的特性,提出了深度数据视点合成优化的模式判决函数。论文的实验部分对基于HEVC的3D视频编码方案3D-HEVC进行系统的实验。首先比较和3D-HEVC和同播(simulcast)技术的性能,其次对基于深度图的3D-HEVC纹理图快速模式判决算法进行实验验证,在保证视频质量的前提下,纹理图编码在原编码方案基础上实现时间上29.3%的精简。最后,在Linux平台下,利用QT和H.265/HEVC标准参考代码,开发了H.265/3D-HEVC的码流分析软件,实现了H.265和3D-HEVC码流的解码、回播、分量显示和基本的码流信息显示。
[Abstract]:With the development of 3D video technology, such as 3D film, 3D TV broadcast and naked eye 3D display technology, 3D video technology has become a research hotspot in academia and industry. Since 2001, the International Organization for Standardization (ISO) expert Group on Motion Images, MPEG, has launched the long-term goal of developing free-view television (Free-viewpoint Television, FTV), which allows 3D video to be viewed with a free change of view. Multi-view video coding (Multi-view Video Coding, MVC) is the first stage of FTV, and 3D video 3DV is the second stage of FTV standard. VCEG and MPEG, the 16th working group of the International Telecommunication Union (ITU), jointly launched the JCT-3V working Group to develop a new generation of 3D video coding technology, in which 3D video coding based on the new generation of efficient video coding technology H.265/HEVC is extended to 3D-HEVC. In 3D-HEVC framework, 3D video is represented as multi-view video plus depth information (Multi-view Video plus Depth, MVD). The coding end adds a new coding tool to the independent view and depth information based on H.265/HEVC. In this paper, the 3D-HEVC scheme of 3D video coding based on HEVC is studied. Firstly, the paper introduces the basic framework and key technology of 3D video coding scheme 3D-HEVC based on HEVC in detail, and introduces three important research difficulties in 3D video coding technology: texture image coding, depth map coding and depth estimation. Secondly, based on the framework of 3D-HEVC, a fast pattern decision algorithm based on depth map is proposed. The algorithm uses the relationship between depth map and texture image in 3D video, divides the corresponding texture image into foreground, main scene and background according to depth map, and simplifies the pattern decision of the corresponding part. The relationship between motion vector and depth value is introduced to select the mode between and within the viewpoint. Secondly, the depth map coding in 3D-HEVC is introduced emphatically. According to the characteristics of large range smooth region and sharp edge of depth map, four new prediction modes are proposed in the intra prediction mode of depth map. According to the characteristics of depth map aided virtual viewpoint synthesis, a pattern decision function for depth data view synthesis optimization is proposed. In the experiment part of this paper, 3D-HEVC, a 3D video coding scheme based on HEVC, is systematically experimented. First, the performance of 3D-HEVC and co-cast (simulcast) is compared, and then the fast pattern decision algorithm of 3D-HEVC texture graph based on depth map is verified by experiments, and the video quality is guaranteed. Based on the original coding scheme, texture image coding achieves 29.3% reduction in time. Finally, using QT and H.265/HEVC standard reference code, we develop the code stream analysis software of H.265/3D-HEVC on Linux platform, and realize the decoding, playback, component display and basic code stream information display of H.265 and 3D-HEVC bitstream.
【学位授予单位】：北京邮电大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN919.81

【共引文献】