基于HEVC的深度视频编码研究

发布时间：2018-04-11 08:32

本文选题：深度视频编码 + 区域　；参考：《天津大学》2014年硕士论文

【摘要】：深度视频在基于多视点加深度(MVD)格式的立体视频系统中起着重要的作用,在编码彩色视频之外编码对应的深度视频可以使立体视频系统利用基于深度的虚拟视点绘制技术(DIBR)生成任意视点的视频,从而以较小的处理代价高效地向观看者提供立体感知。然而,视频编码过程给深度视频带来的噪声会造成虚拟视点质量的下降,需要结合其特性找到一种新的深度视频编码方法。此外由于MVD中的深度视频与对应的彩色视频采集于同一场景,具有高度的相关性。因此,深度视频可以利用已编码的彩色信息辅助编码。本文深入研究了深度图的特性,提出了一种针对深度视频编码的基于区域的三边滤波器来提升编码质量。首先将深度图分成多个区域并对每个区域赋予一个分割标签。然后,为了达到去除噪声并保持边缘尖锐性的目的,每个像素利用与其属于同一区域的临近像素进行滤波。在HEVC平台下对所提出的方法采用3D视频序列进行了验证,实验结果表明视频质量在同比特率下有了一定的提升。在利用彩色视频和深度之间的相关性方面,本文提出一种联合彩色信息的多视点深度视频视差矢量预测方法。该方法首先获得当前编码深度块对应位置的彩色块,提取彩色块预测得到的视差矢量作为深度块的一个候选视差矢量。然后,利用深度块信息计算得到另一个候选视差矢量。最后,对两个候选视差矢量通过计算下采样均方差,选择更精确的一个作为当前深度块的视差矢量预测值。在3D-HEVC编码平台下进行了验证,实验结果表明所提出的方法提高了编码性能。
[Abstract]:Depth video plays an important role in stereo video system based on multi-view and depth MVD format.Coding the corresponding depth video in addition to the coded color video enables the stereo video system to generate arbitrary view video using the depth-based virtual viewpoint rendering technique, thus providing stereo perception to the viewer efficiently at a lower processing cost.However, the noise brought by the video coding process will lead to the degradation of the virtual view quality, so we need to find a new depth video coding method combined with its characteristics.In addition, the depth video in MVD is highly correlated with the corresponding color video in the same scene.Thus, depth video can be encoded using encoded color information.In this paper, the characteristics of depth map are deeply studied, and a region-based trilateral filter for depth video coding is proposed to improve the coding quality.First, the depth map is divided into multiple regions and each region is given a segmentation label.Then, in order to remove noise and maintain edge sharpness, each pixel is filtered by adjacent pixels belonging to the same region.The proposed method is verified by 3D video sequences on the HEVC platform. The experimental results show that the video quality has been improved at the same bit rate.In terms of the correlation between color video and depth, this paper proposes a multi-view depth video disparity vector prediction method combined with color information.Firstly, the color block corresponding to the current coded depth block is obtained, and the parallax vector predicted by the color block is extracted as a candidate disparity vector of the depth block.Then, another candidate parallax vector is obtained by using depth block information.Finally, two candidate parallax vectors are selected as the prediction value of the disparity vector of the current depth block by calculating the sampled mean square error (RMS) of the two candidate parallax vectors.The experimental results show that the proposed method improves the coding performance.
【学位授予单位】：天津大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN919.81

【参考文献】