摘要
计算成像是一种通过联合光学系统和图像处理来实现特定成像功能的新兴研究领域,长期以来,计算成像中的光学与算法联合都采取的是顺序设计模式,即光学系统与图像处理各自分开设计,但这样难以全面发挥二者协同的优势。近年来,随着深度学习技术的飞速发展,基于深度学习架构的光学系统与图像处理算法端到端协同设计方法开启了解决这一问题的大门。一方面,端到端协同设计通过全面探索整个解空间,可以实现光学与图像处理的自动最佳协同;另一方面,端到端协同设计更使得研制基于任务的最优成像系统成为可能。本文首先介绍了基于光学系统与图像处理端到端协同设计框架的进展,然后介绍了我们基于这一框架在平面透镜的宽谱成像、平面透镜的大视场成像、大景深成像、超分辨成像和快照式光谱成像方面的研究进展。
Computational imaging is an emerging research field that combines optical systems and image processing to achieve specific imaging features.For a long time,this kind of combination in computational imaging has adopted a sequential design mode,that is,the optical system and image processing are designed separately,making it difficult to exploit the synergistic advantages of both entirely.With the rapid development of deep learning,the end-to-end co-design method of optics and image processing algorithms based on deep learning architecture has opened the door to solving this problem.On the one hand,end-to-end co-design can realise the automatic optimal collaboration of optics and image processing by comprehensively exploring the entire solution space.On the other hand,end-to-end design makes it possible to develop specific optical imaging systems.In this paper,we review recent advances in end-toend co-design of optics and image processing,including full-colour imaging of flat lenses,large field of view imaging,extended depth of field imaging,and super-resolution imaging,and snapshot spectral imaging.
作者
顿雄
张健
冯诗淇
罗洪
王占山
程鑫彬
DUN Xiong;ZHANG Jian;FENG Shiqi;LUO Hong;WANG Zhanshan;CHENG Xinbin(Institute of Precision Optical Engineering,MOE Key Laboratory of Advanced Micro-Structured Materials,Shanghai Frontiers Science Center of Digital Optics,Shanghai Professional Technical Service Platform for Full-Spectrum and High-Performance Optical Thin Film Devices and Applications,School of Physics Science and Engineering,Tongji University,Shanghai 200092,China)
出处
《光学精密工程》
EI
CAS
CSCD
北大核心
2022年第21期2827-2838,共12页
Optics and Precision Engineering
基金
上海市浦江人才计划资助项目(No.20PJ1414200)
国家自然科学基金资助项目(No.62105243)。
关键词
计算成像
端到端设计
基于任务的成像系统
computational imaging
end-to-end design
task domain imaging system
