多纹理多深度的3D视频码率控制研究

项目名称： 多纹理多深度的3D视频码率控制研究

项目编号： No.61502357

项目类型： 青年科学基金项目

立项/批准年度： 2016

项目学科： 计算机科学学科

项目作者： 邓慧萍

作者单位： 武汉科技大学

项目金额： 21万元

中文摘要： 3D视频的码率控制是3D视频走向网络化应用的重要研究课题。基于多纹理多深度的3D视频数据包含多路视频信号，各类数据特性不同且相互影响，给码率控制带来挑战。本项目从纹理/深度级、帧级、宏块级开展3D视频的码率控制研究，通过分析纹理视频数据、深度信息的特征及相互影响，指导多纹理/多深度的码率分配以保证输出3D视频质量最优；通过分析3D视频码率的影响机理与因素，建立码率控制模型，调整编码参数以达到实际编码码率与目标分配码率一致。在纹理/深度层，建立边缘区域与非边缘区域的虚拟视点失真模型，提出基于立体视频感知失真度量的联合码率分配策略；在帧层，全局优化考虑图像复杂度、位置重要性、左右视点间的参考性，建立目标码率计算模型，改进率失真R-Q模型；在宏块层，利用立体视觉掩蔽特性，建立视觉敏感度失真因子，调整目标码率。通过研究3D视频码率控制的基础理论和技术问题，以推动网络3D视频应用的研究与发展。

中文关键词： 深度图；虚拟视点合成；码率控制；视觉敏感度

英文摘要： Rate control is a key technique for network 3D video application. 3D video data based on multi-view video and depth maps(MVD) include multiple texture and depth incorporate with different characteristics interacting with each other. The problems including how to allocate the depth/texture bitrates to achieve an optimal 3D video quality in the theory, what parameters influence the bitrates of 3D video coding, and how to adjust coding parameters to achieve the accordance with target bit-rate and coding bit-rate, are leading issues. This project research MVD rate control on three levels, namely joint texture/depth bit allocation, frame level rate control and macroblock level rate control.Firstly,we propose a joint texture/depth bit-rate allocation by considering the binocular visual characteristics. We analyze the influence of depth map compression from the synthesized virtual view, and propose a perceptual synthesized view distortion measurement to capture the effect of depth map distortion from the synthesized view. The joint bit-rate allocation problem is transformed into a multi-objective optimization problem, guiding the joint depth/texture bit-rate allocation to achieve the most excellent 3D video quality.Secondly,according to characteristics of stereoscopic video, by taking image complexity, position significance and reference property between the left and right pictures into account, a frame lever rate control algorithm for 3D video coding is presented.Thirdly,we propose a novel macroblock level rate control scheme based on binocular just noticeable difference (BJND) model. In the macroblock layer, visual weighting factor, which is measured by the BJND value of macroblock, is used to adjust the macroblock level bit allocation, so that the rate control results in line with the human visual characteristics. The research has important scientific significance and practical value in video communications and 3D video application. 

英文关键词： depth map;virtual view synthesis;rate control;visual sensitivity