How information is encoded in the visual is vital for understanding human vision. We can use the information in visual scene in two ways; firstly we can find statistical regularities, patterns that underlie the image. Secondly we can manipulate those patterns in order to probe human perception.
Image statistics – Finding patterns
The visual scene is highly complex but contains a significant amount of redundancy. It is believed that the visual cortex exploits this redundancy to produce a sparse and therefore energy efficient coding of the visual scene. I am interested in using principals from information theory to find regularities in images and thus find energy efficient encodings. By deriving models from the ground up using well defined mathematical principals we can put knowledge of how the visual system processes visual input into a sound theoretical framework.
Manipulating images – Face perception.
Human faces exhibit a large amount of variety however this variety is not unconstrained. Statistical models can encompass changes over time such as age, or change features of the face such as femininity/masculinity leadership, health. Starting out as two-dimensional photographs, these models are constantly being improved, with three-dimensional model becoming more common. Collection of face shape (and colour) data still faces numerous challenges particularly in longitudinal studies Three-dimensional scanners are resent innovation longitudinal three-dimensional data has yet to be gathered and photographs are often un-calibrated and taking is varying conditions. My interests in this area are primarily aimed at overcoming these challenges.