Realistic Computer Vision (FINLAYSONG_U26CMP)

About the Project

Primary supervisor - Prof Graham Finlayson

Many algorithms for solving problems in computer vision take an ML approach including using generative AI. As an example, in scenes affected by haze there are many algorithms for seeing through the haze (obviously, very useful in domains such as automated driving). However, the outputs generated by the ML solutions are not realistic [1]: they are different from how the scene actually looked in non-hazy conditions.

In this project we are interested in developing algorithms for solving computer vision problems where the outputs of our algorithms are realistic. Topics of interest include relighting images to discount strong directional light [2] and processing underwater footage to account for signal degradation due to the light-scattering due to the water [3]. The project will involve using/adapting AI algorithms to enforce realism but will also deploy classical processing methods. We will assess the plausibility of our algorithm’s outputs objectively and subjectively.

The PhD student will work as part of the Colour & Imaging Lab (C&IL) currently comprising 10 PhDs+postdocs. This project is in collaboration with the Computer Vision Center, Universite Autonoma Barcelona and the PhD student will have an opportunity to visit there for a period of 3 months. The C&IL has strong links with industry - as partners in projects, hosts for our interns and as job destinations for our graduates – including, Apple, ARM, Google, Hewlett Packard and Meta.

The School of Computing Sciences provides a vibrant research environment for conducting Computing and allied research and training.

Entry requirements

The standard minimum entry requirement is 2:1 or a Masters degree qualification in computer science, mathematics, physics, psychology (psychophysics) and allied numerate disciplines.

Mode of study

Full-time

Start date

1 October 2026