I have expereince carrying out research in several areas relating
to vision and machine learning:
- Multi-Frame Image Super-Resolution — combining several photos of the same scene to create a high-resolution copy, e.g. in order to read text or recognise elements not visible in any one single input image. My work focused on the problems of selecting apropriate image priors, and on how best to handle the problem of registration uncertainty in a principled manner.
- Continuity Errors — My first post-doc position involved having some fun with Prof. Andrew Zisserman, detecting possible continuity errors (e.g. props moving in the background when they shouldn't) in movies.
- Arrow of Time — My current research deals with what aspects of the direction of the flow of time are most readily available to us in videos. We're familiar with the fact that entropy always increases in a closed physical system, but how does that really translate to the visible world around us?
- I spent three years with the Virtual Reality Group at the University of Reading looking at the human basis for 3D visual perception. My research there involved asking human participants to perform simple navigation tasks in virtual reality, then examining the errors they made in reaching the goal points. I created models based on alternative hypotheses about how humans perceive space and remember locations — they use 3D maps in their brains, or they rely on a simple set of "visual features" — and examined how well the observed data aligned with each of the competing hypotheses.
- My under-graduate research project, where I first learned how to be a researcher, dealt with the machine learning problem of detecting "left" and "right" motion planning in the brain from a very basic set-up of three electrodes on a human volunteer's head. The aim of this was to be able to move a computer cursor without the need for actual physical movements, since the signals we were trying to detect could be created by imagining motions that were never physically made.
Papers relating to most of this work can be found in the publications section of the site.