【作者】 韩雪；

【导师】 冯桂；

【作者基本信息】 华侨大学 ， 信息与通信工程， 2019， 硕士

【摘要】 随着互联网的普及,人们对视频清晰度的要求越来越高,如何在有限的带宽资源下高效的传输视频信息是多媒体技术中亟待解决的问题。为了高效编码和传输3D视频,基于编码2D视频的高性能视频编码标准(High Efficiency Video Coding,HEVC),联合视频编码小组提出了新一代3D视频编码标准3D-HEVC(3D High Efficiency Video Coding)。3D-HEVC加入多种新技术并有效减少了编码失真,但计算复杂度依然较高,因此,提高3D视频的编码效率,具有重要意义。本文首先介绍了3D-HEVC采用的编码框架并对关键技术做了详细说明,基于3D视频的时域、空域和视点间冗余性,结合图像的纹理特征,分别对帧内预测和帧间预测过程做优化。第一,提出了基于Otsu’s算子的深度图帧内预测快速编码算法。首先,统计深度图帧内预测的结果,分析编码块(Coding Uint,CU)纹理复杂度与编码结果的关系。用Otsu’s算子计算当前CU的最大类间方差值,判断当前CU是否平坦,对平坦CU终止四叉树分割并减少帧内模式检测的数目。根据子CU与上一层CU的相似性,利用已编码的上一层CU对提前终止分割算法做优化。本算法与3D-HEVC的原始算法相比,减少40.1%的编码时间,而合成视点的质量几乎无变化。第二,提出了深度图帧间预测快速编码算法。深度图中大部分CU采用较大编码尺寸和Merge模式,采用较小尺寸和非Merge模式的CU集中分布在前景区域,其运动特征较复杂。根据灰度值反映了物体相对距离这一特征,利用灰度值把深度图分为背景区域,前景区域和复杂区域,分别执行不同CU尺寸决策过程。根据纹理图与深度图表示相同场景,利用纹理图运动特征提前判决深度图中运动复杂的区域。分析三类区域的模式选择结果,根据母CU的运动矢量特征提前判决子CU的编码模式。实验结果表明,本算法的编码时间减少35.62%,合成视点平均比特率增加0.27%。更多还原

【Abstract】 With rapid development of the internet industry,people’s needs for Hi-Definition picture and video are growing.How to transmit huge amount of data through limited bandwidth are key issues in multimedia technology.In order to encode 3D video efficiently,the joint collaborative team on video coding proposed a coding standard for 3D video based on traditional 2D video coding standard HEVC,names 3D High Efficiency Video Coding(HEVC).3D-HEVC adopted coding structure and most coding methods in HEVC standard,it also added several new methods to improve coding efficiency.On one hand these new methods reduce the quality loss,on the other hand,it increases computational burden.In this paper,we introduce the encoding structure and the coding process of 3DHEVC standard.We analyse the redundancy in spatial,temporal and inter-view domain,texture complexity are also considered.At last,we proposed two fast algorithms for depth map intra prediction and inter prediction process.For intra coding process,we propose a fast algorithm based on Otsu’s method.Firstly,analyze the relationship between texture complexity of depth CUs and quadtree splitting results.Applying Otsu’s method to each depth level to get the maximum inter class variance,and detect smooth CUs.Stop further splitting process of smooth CUs and simplify the intra prediction process.We further optimize our algorithm by using rate distortion cost based on the correlation between different layers of CUs.Experimental results show that our proposed algorithm can save about 40.1% of time consumption on average with negligible loss of coding performance.For inter coding process,we proposed a fast algorithm based on background segmentation method.Most of CUs use Merge mode and encode with large sizes,only CUs with complex motion features split into depth level 2 and 3,these CUs are at front regions.Depth value represent the absolute distance between objects and cameras.We classify each CU into background region,non-background region through depth value.Non-background region can be further classified into edge region and smooth region by checking whether it contains edges.The total value of transformed block in texture map are used in our algorithm to simplify quad-tree splitting process for each type of CUs.For fast mode decision process,obtain the motion information of already coded parent CUs to reduce the number of modes need to be checked in inter coding process.Our algorithm is implemented on HTM16.0,it achieves an average time reduction of 35.62%,with bitrate loss of 0.27% for synthesized views.In the end,we summarize the research on 3D-HEVC and propose the research direction in the future.更多还原