CAREER: Optimized Sensing and Recovery for Computational Imaging

在过去的几十年里，相机已经发生了巨大的变化，因为它们已经变得紧凑并且广泛用于移动的设备中。尽管如此，仍然存在巨大的改进机会，特别是随着社区超越消费者摄影，并开始在严格的限制下构建用于感知和理解新环境的相机。这些限制可能是由于对组件的大小、形状或重量的物理要求，捕获和处理数据所需的时间，或整个系统的成本和能量阈值而产生的。本研究的重点是开发新的方法来感知和处理视觉信息，同时考虑到真实的世界中的数据，传感器，算法和任务的约束和要求。该研究的成功成果将使消费者摄影、机器视觉和自动化以及科学/医学成像等多种应用受益。通过教育课程和研究经验培训不同群体的本科生和研究生是该项目的一个组成部分。该项目的教育和推广部分涉及通过校园年度研究日向K-12学生和教师传播计算成像研究，以及通过与摄影博物馆合作向公众传播。计算成像提供了一个通用的框架来共同设计传感硬件和计算软件，以构建新颖和非传统的相机。这项研究将解决一些基本的理论和算法问题，优化传感，表示和恢复计算成像。该研究被组织成三个相互关联的推力：（1）优化和扩展计算成像系统可以使用可编程光学器件真实捕获的测量空间。(2)以端到端的方式学习感知、数据表示和恢复算法。(3)为计算成像系统开发有效的算法，超越线性和平移不变模型。从这项研究中获得的见解将帮助我们了解现实计算成像系统中存在的一些基本限制，并提供在给定约束条件下优化它们的工具。本研究开发的模型和算法将通过真实的成像实验进行验证。该奖项反映了NSF的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Cameras have evolved tremendously over the past few decades as they have become compact and widely available in mobile devices. Nevertheless, vast opportunities for improvements still exist, especially as the community moves beyond consumer photography and start building cameras for sensing and understanding new environments under strict constraints. These constraints can arise because of physical requirements on the size, shape, or weight of the components, time required to capture and process the data, or cost and energy thresholds for the entire system. The focus of this research is to develop novel methods to sense and process visual information while taking into account the constraints and requirements on data, sensors, algorithms, and tasks in the real world. A successful outcome of the research will benefit diverse applications spanning consumer photography, machine vision and automation, and scientific/medical imaging. Training of a diverse group of undergraduate and graduate students through educational courses and research experience is an integral part of this project. The education and outreach component of this project involves dissemination of computational imaging research to K-12 students and teachers through annual research days on campus, as well as to the general public by collaborating with a museum of photography. Computational imaging offers a general framework to co-design sensing hardware and computational software to build novel and unconventional cameras. This research will address a number of fundamental theoretical and algorithmic questions related to optimized sensing, representation, and recovery for computational imaging. The research is organized into three inter-related thrusts: (1) Optimize and expand the space of measurements that computational imaging systems can realistically capture using programmable optics. (2) Learn sensing, data representation, and recovery algorithms in an end-to-end manner. (3) Develop efficient algorithms for computational imaging systems beyond linear and shift-invariant models. The insights gained from this research will help us understand some of the fundamental limits that exist in realistic computational imaging systems and provide tools to optimize them within the given constraints. Models and algorithms developed in this research will be validated with real imaging experiments.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.