计算成像——全光视觉信息的设计获取

Computational Imaging——Design for Acquisition of Plenoptic Visual Information

真实世界的视觉光信息是复杂的高维度连续的信号;而现有的数字成像方式是低维耦合离散采集,在成像的各个维度——空间分辨率、时间分辨率、视角及深度、颜色(光谱)等均已达到瓶颈:这极大限制了场景视觉信息的全面获取。计算成像,综合信号处理、光学、视觉、图形学等多学科知识,可突破经典成像模型和相机硬件的局限,以更加全面、精确地捕捉真实世界的视觉信息。文章从计算成像的进化史、计算成像的基本原理、计算成像研究动态、计算成像的关键技术以及计算成像的未来展望等方面,从全光视觉信息的设计获取角度阐述了计算成像的相关概念与理论。

Visual light information in the real world is a complex, high-dimensional and continuous signal. However the existing digital imaging method is low-dimensional coupled discrete acquisition, which has reached the bottleneck in every dimension of imaging, such as spatial resolution, temporal resolution, visual angle and depth, color(spectrum), thus greatly limiting the full range of scene visual information. Computational imaging, which integrates multi-disciplinary knowledge in cluding signal processing, optics, vision, graphics and so on, can break through the limitations of classical imaging model and camera hardware to capture real world visual information more comprehensively and accurately. In this paper, the related concepts and theories of computational imaging are elaborated from the aspects of evolutionary history, basic principles, research trends, key technologies and future prospects.