){kind=small}
Current PhD Research
Analysing Videos of Fish in the Field
Participants
Summary
Mazvydas, as part of the Colour & Imaging Lab at the University of East Anglia is developing automated video analysis techniques for fishing industry applications. The project initially focused on analyzing conveyor belt footage from fishing vessels to classify and measure fish, a process traditionally done manually. The new phase of research aims to extend these techniques to more challenging and variable environments, like outdoor fish markets or onboard small fishing vessels, where lighting conditions are inconsistent and videos are often captured on handheld devices like mobile phones. The main challenge is adapting these techniques to work well in different lighting and movement conditions. To achieve this, the lab is exploring advanced technologies in image processing, machine learning, and 3D computer vision. Initially, the project uses synthetic data for algorithm development, gradually incorporating real-world data annotated by experts. This research aims to create efficient, accurate, and adaptable video analysis tools for automation of the fishing industry.
)
Understanding and mitigating the problem of highlights in remote sensing with application to coastal surveying
Participants
Summary
My PhD research focuses on enhancing vision systems drone imagery over coastal areas by mitigating the effects of sun glare. These images, used by Cefas—my industrial partner—for environmental classification, often suffer from clipped highlights, leading to inaccurate identification of biological and physical features. To address this, we are developing a novel Retinex algorithm. Our method processes high dynamic range images generated from multiple short-exposure shots. This captures a wider dynamic range and reduces the probability of clipped areas where information is lost. The Retinex separates and removes illumination effects, preserving the true reflectance of the scene. This enables it to recover details lost in bright regions and maintain perceptual colour consistency under varying lighting conditions similar to the human eye. Compared to traditional Retinex methods, our approach offers faster convergence, compatibility with the human vision system, improved visual quality, and reduced artefacts. Unlike deep learning solutions, it requires no training and generalizes well across diverse image datasets.
PGR Poster (PDF, 667KB)
Partners
)
){kind=logo}
Improving Dementia Diagnosis Through Cortical-Subcortical Brain Region Analysis
Participants
Summary
Based at the Colour and Imaging Lab at the University of East Anglia, and part of Dr. Sami's TENSOR Lab at the Norwich Medical School, Samuel investigates advanced computational neuroscience methods to study the early stages of Alzheimer’s Disease (AD) pathology. His cross-disciplinary research primarily focuses on the connectivity between cortical and subcortical networks via various MRI analysis techniques. By combining anatomical, structural diffusion, and functional resting state MRI data with detailed cellular, neurotransmitter, and receptor profiles, he aims to uncover the underlying mechanisms of neurodegeneration.
In one of his current projects, Samuel is enhancing the predictive accuracy of the brain-Predicted Age Difference (brain-PAD) metric using an ensemble, region-focused approach. This work explores the relationship between structural brain-PAD and cognitive reserve, with the goal of producing robust tools suited for memory clinics. Additional ongoing projects involve evaluation of thalamocortical projections to detect subtle changes in functional connectivity during preclinical AD. He also plans to investigate hippocampal-cortical white matter connectivity using high-density tractography approaches, building on his previous work using graph-based connectome analysis methods.
PGR Poster (PDF, 0.99KB)
)
Partners
){kind=logo}
){kind=logo}
Incorporating Physical Constraints for Improving Spectral Recovery
Participants
Summary
My PhD research explores how to enhance digital imaging by leveraging multispectral cameras—devices that capture detailed spectral information beyond the standard three RGB channels. While multispectral cameras are increasingly valued for their ability to measure light with greater precision, they typically suffer from low spatial resolution. To address this, I investigate RGB-guided pan-sharpening techniques that fuse high-resolution RGB images with low-resolution multispectral data. This allows us to produce images that are both spatially sharp and spectrally rich. Rather than focusing solely on reconstructing full multispectral images, I also explore how MSC data can improve RGB cameras for better colour representations. My work contributes to making advanced imaging systems more accessible and practical, with potential applications in photography, cultural heritage, and medical imaging.
)
Funding
){kind=logo}