3D-HEVC虚拟视点合成优化及预测技术研究

发布时间：2018-02-11 18:51

本文关键词： HEVC 3D-HEVC 虚拟视点合成视点合成预测视点合成优化　出处：《西安电子科技大学》2014年硕士论文　论文类型：学位论文

【摘要】：三维视频(3DV)可提供观看场景的深度信息,为用户提供立体视觉体验。近期运动图像专家组(MPEG)以多视点视频加深度(MVD)作为3DV数据格式,在HEVC基础上制定新一代3DV编码标准3D-HEVC。相比于传统的二维视频,MVD增加了额外视点的视频信息和深度图信息,使得编码码率大大增加。由于不同视点视频取自同一拍摄场景,视点间具有较强的相关性,可用于提高编码效率。同时,由于深度信息主要用于解码端的虚拟视点合成而非直接观看,因此其率失真优化准则与传统视频编码有所不同。本文主要研究了3D-HEVC中纹理编码的视点合成预测技术和深度编码的视点合成优化技术。视点合成预测是一种特殊的视点间预测编码方法,其利用视点间匹配块和深度图合成出当前位置的预测块进行预测编码。将虚拟深度块分为4×8或8×4子块,然后在每个子块中比较左上、右上、左下和右下位置的深度值选择最大深度值。由于3D-HEVC中深度图不做去方块滤波,在选择最大深度值时会受到块效应的影响。基于此,本文提出两种改进方法,第一种取每个子块中间位置的深度值作为最大深度值,第二种是取每个子块四个角向内偏移一个像素的四个位置求最大值以有效避开边界的影响。实验结果表明,本文所提两种方法使非基本视点的码率平均降低了0.07%和0.18%。且第二种方法选取的最大深度值对应的视差向量更能准确预测当前块。不同于纹理图,深度图其编码目的主要用于解码端合成虚拟视点。因此,现有3D-HEVC提出了基于虚拟视点质量最优的深度编码算法。编码中间深度需要计算6个合成虚拟位置视点的失真,目前率失真准则中6个视点失真采用等权值机制。由于0.25、1.25和0.75、1.75位置分别距离左边和右边视点比较近,失真比较小;而中间位置0.5和1.5相对失真比较大,所以赋予这两个失真较大权值,以此更能精确选择出失真最小的编码模式。本文尝试设计了11组不同权值进行了实验测试。初步实验结果显示,在较大QP下增大0.5及1.5位置的权重可使得整体合成质量增加;每个QP下选择出一组效果较优的权值,可使合成视点质量平均提高0.05%。通过大量实验,本文拟合出了不同QP与较优权重的关系以提高合成视点质量。
[Abstract]:3D Video V3 DVV) can provide the depth information of the viewing scene and provide the stereoscopic vision experience for the user. The recent moving Image expert Group (MPEGG) uses multi-view video plus depth MVDD as the 3DV data format. A new generation of 3DV coding standard 3D-HEVCis developed on the basis of HEVC. Compared with the traditional 2D video, the coding rate of the new 3DV coding standard is greatly increased because the video information and depth map information of the additional viewpoint are added to the traditional 3D video coding standard. Because different view video is taken from the same shooting scene, the coding rate of the new 3DV coding standard is greatly increased. Because the depth information is mainly used in the virtual view synthesis of the decoder, it can be used to improve the coding efficiency. Therefore, the rate-distortion optimization criterion is different from traditional video coding. This paper mainly studies the viewpoint synthesis prediction technology in 3D-HEVC texture coding and the viewpoint synthesis optimization technology in depth coding. Viewpoint synthesis prediction is a special method. Interview prediction coding method, The prediction block of the current position is synthesized by using the viewpoint matching block and the depth map. The virtual depth block is divided into 4 脳 8 or 8 脳 4 subblocks, and then the upper left and the upper right are compared in each sub-block. The depth values of the left and right lower position choose the maximum depth value. Because the depth map in 3D-HEVC is not filtered by block, it will be affected by block effect when selecting the maximum depth value. Based on this, two improved methods are proposed in this paper. The first is to take the depth value of the middle position of each sub-block as the maximum depth value, the second is to obtain the maximum value by taking the four positions of one pixel offset one pixel from the four angles of each sub-block in order to avoid the influence of the boundary effectively. The two methods proposed in this paper reduce the average bit-rate of non-basic viewpoint by 0.07% and 0.18.The parallax vector corresponding to the maximum depth value of the second method can accurately predict the current block. The coding purpose of depth map is mainly used for synthesizing virtual viewpoint at decoding end. Therefore, 3D-HEVC has proposed a depth coding algorithm based on the optimal quality of virtual view. The intermediate depth of encoding needs to calculate the distortion of the view of six synthetic virtual positions. In the current rate-distortion criterion, six view distortion uses the equal weight mechanism. Because the positions of 0.2525 and 0.75U 1.75 are relatively close to the left and right view, respectively, the distortion is smaller, while the relative distortion of the middle position 0.5 and 1.5 is relatively large. Therefore, these two distortion weights can be given to select the coding mode with minimum distortion more accurately. In this paper, 11 groups of different weights have been designed for experimental test. The preliminary experimental results show that, Increasing the weight of 0.5 and 1.5 positions under the larger QP can increase the overall synthetic mass. Under each QP, a group of better weight values can be selected, and the average weight of the composite viewpoint can be increased by 0.05. Through a large number of experiments, In this paper, the relationship between different QP and better weight is fitted to improve the quality of composite viewpoint.
【学位授予单位】：西安电子科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN919.81

【相似文献】