Computational imaging with the human brain

Wang, Gao (2024) Computational imaging with the human brain. PhD thesis, University of Glasgow.

Abstract

Human augmentation, which involves enhancing cognitive and physical abilities as a natural extension of the human body, has been significantly advanced by Brain-Computer Interfaces (BCIs). This thesis explores BCI-based human augmentation, focusing on computational ghost imaging and developing a phenomenological brain model for Steady-State Visual Evoked Potentials (SSVEPs).

Initially, the concept of BCIs as a conduit for computational imaging is introduced, demonstrating the potential to integrate brain function with external silicon-based processing systems. A key example is the ghost imaging of a hidden scene using the human visual system in conjunction with an adaptive computational imaging scheme. This technique, known as projection pattern ‘carving,’ utilizes real-time brain feedback to modify light projector patterns, resulting in more efficient and higher-resolution imaging. This brain-computer connectivity represents a form of augmented human computation, potentially expanding the sensing range of human vision and offering new methodologies for studying the neurophysics of human perception. An illustrative experiment is presented, highlighting how image reconstruction quality can be influenced by simultaneous conscious processing and readout of perceived light intensities.

Subsequently, the thesis delves into the phenomenon of SSVEP, which has attracted attention across various fields including neuroscience and human augmentation. The analysis of SSVEP under multiple frequency stimuli, a complex task due to frequency intermodulation terms, is addressed by proposing a phenomenological model. This model provides a mathematical framework for analysing the essential frequency mixing features in SSVEP when exposed to multifrequency stimuli. The analysis is extended to both narrowband and broadband categories using analytical and statistical methods. Experimental results confirm the model’s accuracy, shedding light on the mathematical model behind SSVEP responses to multiple frequency stimuli and offering insights for practical applications and a deeper understanding of this phenomenon.

Addressing the neuromorphic aspect of SSVEP, the thesis discusses the extensive use of SSVEP in BCIs due to their stability and efficiency in connecting the computer and the brain using simple flickering light. Moving beyond prior research that focused on low-density frequency division multiplexing techniques, this work demonstrates the feasibility of efficiently encoding information in SSVEPs through high-density frequency division multiplexing, involving hundreds of frequencies. The capability to transmit complete images from the computer to the brain/EEG read-out within a short timeframe is also illustrated. High-density frequency multiplexing enables the implementation of a photonic neural network that leverages SSVEPs for performing simple classification tasks, showcasing promising scalability through serial brain connectivity. This research opens innovative pathways in neural interfacing, with implications for assistive technologies and cognitive enhancement, significantly advancing human-machine interaction.

Lastly, the concept of SSVEPs is extended to multi-frequency light modulation, relying on the broadband scenario of the phenomenological brain model. The research demonstrates the brain’s ability to support the SSVEP read-out transmitting image. When the bandwidth spans more than an octave, the higher harmonics and nonlinear mixing between signal pairs overlap with the fundamental harmonics, creating a highly complex EEG signal. By utilizing a DNN trained on synthetic data, it is feasible to retrieve the original input signal, which can be employed to reconstruct images with each pixel encoded at a distinct single frequency. This approach facilitates precise image transmission, with each pixel encoded at a unique frequency. The BCI developed in this thesis enables multi-channel data transmission, and networked interfaces, and has potential applications in diagnostics, assistive technologies, and cognitive enhancement tools.

• Item Type: Thesis (PhD)
• Qualification Level: Doctoral
• Additional Information: Supported by funding from the China Scholarship Council (CSC).
• Colleges/Schools: College of Science and Engineering > School of Physics and Astronomy
• Supervisor's Name: Faccio, Professor Daniele and Zhao, Dr. Hubin
• Date of Award: 2024
• Depositing User: Theses Team
• Unique ID: glathesis:2024-84826
• Copyright: Copyright of this thesis is held by the author.
• Date Deposited: 17 Jan 2025 16:23
• Last Modified: 17 Jan 2025 16:43
• Thesis DOI: 10.5525/gla.thesis.84826
• URI: https://theses.gla.ac.uk/id/eprint/84826

Related URLs

• Article DOI
• Article DOI
• Article DOI
• Article DOI
• Article DOI
• Article DOI
• Conference proceeding