First Advisor

Atul Ingle

Term of Graduation

Spring 2024

Date of Publication

6-7-2024

Document Type

Thesis

Degree Name

Master of Science (M.S.) in Computer Science

Department

Computer Science

Language

English

Subjects

Computational imaging, Resource-constrained imaging, Single-Photon Cameras, Single-photon imaging, SPAD

Physical Description

1 online resource (x, 64 pages)

Abstract

The modern world is built of images. However, in our goal to photograph and replicate what the human eye is capable of seeing, we are throttled by the restrictions of conventional imaging sensors in high- and low-illumination environments. Single-photon cameras (SPCs) have recently emerged as a promising alternative to conventional camera sensors for capturing images in challenging conditions such as high-dynamic range and fast scene motion. Compared to traditional CMOS cameras, SPCs exploit the arrival of individual photons rather than using an aggregate photon count to compute the brightness of pixels. However, SPCs are extremely resource-intensive, making them inconvenient for power-limited applications such as smartphones, mobile devices for augmented and virtual reality, low-power drones, and autonomous robots. In many imaging settings, we are restricted by the volume of photon data that can be stored and processed by individual pixels in large format (megapixel or larger) SPCs.

The thesis of our work is that SPCs can reconstruct high quality images with far fewer photons per pixel than previously thought necessary. To show this, we present an imaging pipeline including a resource-constrained multiplexing capture algorithm and a state-of-the-art deep learning-based denoiser method using transformers. In particular, we address a fundamental question in single-photon imaging---what is the minimum number of photons needed per pixel to construct an accurate image, and how few photons can one receive before an image is no longer recoverable? Our experiments on a large image dataset show that our pixel-sharing and post-processing denoising techniques can improve image quality by up to 5 dB peak-signal-to-noise ratio in low-light settings when compared to baseline capture methods. By allowing groups of pixels to share in-pixel memory and computing resources, our methods will lower the circuit and hardware complexity of future SPCs without compromising image quality.

Rights

In Copyright. URI: http://rightsstatements.org/vocab/InC/1.0/ This Item is protected by copyright and/or related rights. You are free to use this Item in any way that is permitted by the copyright and related rights legislation that applies to your use. For other uses you need to obtain permission from the rights-holder(s).

Persistent Identifier

https://archives.pdx.edu/ds/psu/42322

Available for download on Saturday, June 07, 2025

Share

COinS