Author ORCID Identifier

https://orcid.org/0000-0002-8994-2429

Date of Award

2025

Document Type

Thesis (Ph.D.)

Department or Program

Computer Science

First Advisor

Wojciech Jarosz

Abstract

Many scientific fields rely on the capture and modeling of light to extract underlying information about the world. Often, more information can be extracted by capturing more about the nature of the light, such as its spectral shape or polarization state. While polarization is a relatively unexplored topic in computer graphics, when used in tandem with other recent advancements in the field it has enormous potential to improve both forward and inverse models in other scientific disciplines. We demonstrate this potential in two distinct settings in this thesis.

First, we apply the capture of polarized light to an inverse problem in the setting of computational photography. We propose a novel design for a do-it-yourself hyperspectral imaging system driven by taking multiple photographs through tunable, polarization-induced, spectral filters. These filters can generate a continuous family of broadband transmission spectra via simple rotations of stacked polarizers and waveplates. Our prototype demonstrates that our approach can achieve comparable quality to prior work at reduced cost, while the new design space holds ample opportunity for increased quality and flexibility with professional manufacturing.

Next, we investigate the capture of polarized light in the setting of remote sensing. Existing forward models used by the remote sensing community are typically accurate and fast, but sacrifice flexibility by assuming the atmosphere or ocean is composed of plane-parallel layers that are laterally homogeneous. Monte Carlo forward models, such as those favored by the computer graphics community, can handle more complex scenarios such as 3D spatial heterogeneity, but are relatively slow. We demonstrate that Monte Carlo forward models in computer graphics are capable of sufficient accuracy for remote sensing by extending a forward and inverse modeling framework recently developed in the computer graphics community to simulate simple atmosphere-ocean systems. We show that our framework is capable of achieving error on par with codes currently used by the remote sensing community on benchmark results. Lastly, we demonstrate that our framework can be used to retrieve parameters in a variety of simple inverse problems.

Download

Share

COinS