多深度融合感知的多视点视频联合处理与高效编码

Realistic three-dimensional representation of a scene in the video field of military , aerospace , medical, film and television production has a wide range of applications , depth perception at the core of the visual characteristics of the three-dimensional video processing is driven by the current international research frontier and hot , the mechanism involved in the visual image cross- processing and optimization of multiple disciplines such as the direction , the challenge lies in the multiview depth perception -driven multi-view video combined treatment involving core issues include : time - space - view depending on the consistency of multiple depth perception , binocular depth perception measure, more efficient coding methods like the depth perception . This project attempts to reveal the three-dimensional video signal processing and visual perception collaborative optimization mechanism between treatments, multiview depth research collaborative optimization method , to explore the time - space - geometry and depth of field of visual perception of depth and more consistency problems, to achieve global optimization multiview depth of field ; build multi significant depth perception integration model to explore the depth perception optimized rate allocation methods and technology, depth perception and more efficient multi-view video coding to solve multi- dimensional depth perception driven video processing the key issue, the establishment of multi- depth perception of multi-view video processing combined with efficient coding platform, a breakthrough in the theory and key technology research, promote the wider use as the core of a multi -view 3D video .

现实场景的三维视频表示在军事、航天、医疗、影视制作等领域有着广泛的应用前景，以深度感知为核心的视觉特性驱动的三维视频处理是当前国际研究的前沿与热点，涉及视觉机理、图像处理与优化等多个学科方向的交叉，其挑战在于以多视点深度感知为驱动的多视点视频联合处理，涉及到的核心问题包括：时-空-视多深度感知一致性、双目深度感知测度、多深度感知下的高效编码方法等。本项目力图揭示三维视频在信号处理与视觉感知处理之间的协同优化机理，研究多视点深度协同的优化方法，探索时-空-视多深度场的几何与深度感知一致性难题，实现全局优化的多视点深度场；构建多深度融合感知的显著性计算模型，探索深度感知优化的码率分配方法和技术，实现多深度感知的多视点视频高效编码，解决多深度感知驱动的三维视频处理的关键问题，建立多深度感知的多视点视频联合处理与高效编码平台，在理论和关键技术研究上取得突破，推动以多视点为核心的三维视频的广泛应用。

本项目执行期间，围绕感知的多视点视频处理的关键问题，抓住视觉媒体技术高速发展与广泛应用、视觉信息感知多元化的契机，研究多视点视频中立体视觉的感知机制与编码优化模型。重点研究研究多视点深度协同的优化方法，通过提出跨视点的宏超像素单元及联合优化，实现全局优化的多视点深度场；构建多深度融合感知的显著性计算模型，采用聚类方法，提升深度显著度计算中的鲁棒性；进一步在多视点光场数据上研究深度感知优化的码率分配方法和技术，实现多深度感知的多视点视频高效编码。最终取得核心理论与关键技术上的突破，实现原创性的科研成果，建立感知的多视点三维视频数据高效处理平台，进一步推动三维视频在视觉感知表示、多视联合处理、高效感知编码等方面的研究与发展。