面向超高清晰度的三维高动态范围视频系统视觉质量评价

Three dimensional high dynamic range (3D-HDR) video system with ultra high definition is development direction of the new generation of video system. One of the key technologies of 3D-HDR video system is how to evaluate visual quality of the video during the processes of acquisition, coding, and multiple dynamic range displays, so as to optimize the resource allocation and systematic design. 3D-HDR video's high quality, high definition, high dynamic range characteristics make it different from the existing low dynamic range video in visual perception characteristics. The existence of tone mapping in the processes of encoding and high dynamic range display also makes the video quality assessment of 3D-HDR video to be more complex. Up to now, the research on visual quality assessment of ultra high definition 3D-HDR video system is still in its infancy. In this project, perception experiments will be designed to analyze factors influencing human's 3D visual perception characteristics under high quality and high definition conditions, and the perception characteristics will be modeled. Then, based on the 3D visual perception model, characteristics of 3D-HDR video distortion will be explored, using non-negative matrix factorization, manifold learning, tensor decomposition, dictionary learning and other data analysis tools to achieve accurate quality assessment, and provide effective visual quality assessment theory and methods for video acquiring, encoding, and multi dynamic range display of ultra high definition 3D-HDR video system.

面向超高清晰度的三维高动态范围(3D-HDR)视频系统是新一代视频系统的发展方向。如何有效评价其内容获取、编码、多动态范围显示过程中视觉质量以优化资源分配和系统设计是实现3D-HDR视频系统的关键技术。3D-HDR视频的高质量、高分辨率、高动态范围特点使其在视觉感知特性上有别于现有低动态范围视频；且在编码与多动态范围显示过程中采用了色调映射，使得质量评价问题更为复杂。目前对超高清晰度3D-HDR视频系统视觉质量评价的研究尚属起步阶段。本项目从研究影响人眼关于高或较高质量、高分辨率前提下3D视觉感知特性要素出发，设计3D-HDR感知实验，分析各因素影响，建立模型；基于3D视觉感知模型探索3D-HDR视频失真的特征描述，利用非负矩阵分解、流形学习、张量分解、字典学习等分析工具提取特征实现精确质量评价，为超高清晰度3D-HDR视频获取、编码与多动态范围显示等提供有效的3D视觉质量评价理论与方法。

面向超高清晰度的三维高动态范围视频系统是新一代视频系统的发展方向。而如何评价其视频获取、编码、显示过程中的视觉质量，以优化资源分配，使之与人类视觉感知相适配，从而获得最优的用户视觉感知质量极为重要。三维高动态范围视频的高质量、高分辨率、高动态范围特点使其在视觉感知特性上有别于现有低动态范围视频；且其在编码与多动态范围显示过程中采用了色调映射，使得其质量评价问题更为复杂。.本项目从分析影响人眼关于高质量、高分辨率前提下立体视觉感知特性的要素出发，设计了立体/三维与高动态范围视觉内容的主观感知实验，建立相应的主观测试数据库，分析了各因素对人眼视觉感知特性的影响，为超高分辨率三维高动态范围视频系统中视觉内容的客观质量评价模型设计提供了依据；并据此提出了三维高动态范围视频系统的视觉内容生成优化方法。基于人眼视觉感知特性，探索了三维高动态范围视频系统中视觉内容的失真机理及其特征描述，利用张量分解、非负矩阵分解、流形学习、字典学习等分析工具提取特征，提出了包括高动态范围图像/色调映射图像、三维图像/视频等的客观质量评价模型和方法，为超高分辨率三维高动态范围视频系统视觉内容获取、优化编码与多动态范围显示等提供有效的客观质量评价方法与技术。.本项目在超高分辨率三维高动态范围视频系统的视觉内容质量评价及其应用方面取得了较好的成果，已完成学术论文31篇，发表或录用在IEEE Transactions on Computational Imaging、IEEE Transactions on Image Processing、IEEE Transactions on Circuit System for Video Technology、Future Generation Computer Systems、Signal Processing、Neurocomputing、Journal of Visual Communication and Image Representation、Signal Processing: Image Communication等国际权威/重要SCI期刊，以及国内核心期刊、ICIP、ICASSP、ICME等重要国际会议上；申请和授权发明专利14件；培养博士/硕士毕业生13名。