CAREER: A Signal Processing Framework for Computational Imaging: From Theory to Applications

计算成像（CI）已经成为一种强大的工具，用于设计各种应用中的成像系统，包括消费成像，机器视觉，监控，遥感和显微镜。CI背后的关键思想是，光学、传感器和算法的联合设计和优化迅速扩大了成像的设计空间，实现了迄今为止无法实现的性能特性。该项目开发新型成像系统，通过利用光学和信号处理的协同设计，可以突破尺寸，空间分辨率，时间分辨率，视场，重量和成本的障碍。成像本身具有非常广泛的应用，涵盖消费者/医疗/科学成像，机器视觉和遥感。在推进该项目的科学目标的同时，该项目将计算成像和显示概念整合到莱斯大学提供的几门课程中，并正在为经济困难的高中学生及其教师开发一个关于成像的夏季研讨会，从而提高K-12 STEM的参与度。本研究开发了一个基于高斯混合模型的计算成像信号处理框架（光学，传感器和算法的协同设计）。该模型与相关算法一起沿着，将首次实现（a）现有CI系统的基本性能界限，（B）使设计过程系统化并开发新的CI系统，以及（c）表征和理解CI的操作机制。研究团队通过在三个不同的应用领域创建数量级的性能改进来展示该框架的重要性和实用性：（1）无透镜相机：设计，开发和演示薄的无透镜相机，该相机由一系列放置在裸露的传统传感器顶部的掩模组成，具有相关的解复用算法;（2）飞行中的光成像：利用传统的飞行时间传感器和适当的照明/传感器调制和重建技术，设计、开发和演示超快成像;以及（3）高分辨率、远距离成像：利用相机阵列和基于相位恢复的算法，设计、开发和演示远距离亚衍射成像分辨率。

Computational Imaging (CI) has emerged as a powerful tool for the design of imaging systems over a wide variety of applications, including consumer imaging, machine vision, surveillance, remote sensing, and microscopy. The key idea behind CI is that joint design and optimization of optics, sensors and algorithms rapidly enlarges the design space for imaging enabling performance characteristics that were hitherto unachievable. This project develops novel imaging systems that can break through the barriers of size, spatial resolution, temporal resolution, field of view, weight and cost by leveraging on the co-design of optics and signal processing. Imaging as such, has a very broad range of applications spanning consumer/medical/scientific imaging, machine vision and remote sensing. In conjunction with advancing the scientific goals of this project, the project integrates computational imaging and display concepts into several courses offered at Rice University and is developing a summer workshop on imaging to economically disadvantaged high school students and their teachers, thereby increasing K-12 STEM engagement. This research develops a signal processing framework for Computational Imaging (co-design of optics, sensors and algorithms), that is based on Gaussian Mixture Models. The model along with associated algorithms, will, for the first time, enable (a) fundamental performance bounds for existing CI systems, (b) make the design process systematic and develop novel CI systems, and (c) characterize and understand the regimes of operation for CI. The research team demonstrates the importance and utility of the framework, by creating order of magnitude performance improvements in three different application areas: (1) lens-free cameras: design, develop and demonstrate thin, lens-free cameras that consist of a series of masks laid on top of a bare conventional sensor with associated algorithms for demultiplexing; (2) light-in-flight imaging: design, develop and demonstrate ultra-fast imaging using conventional time-of-flight sensors and appropriate illumination/sensor modulation and reconstruction techniques; and (3) high-res, long-range imaging: design, develop and demonstrate sub-diffraction imaging resolution at long ranges, using a camera array and phase retrieval based algorithms.