Published: 2nd November 2006
Have you ever wondered why we more readily recognise someone of our own race? Or why we can sometimes not tell a person’s gender? Or why we can be unsure if we have seen a face before?
A fascinating research project in Manchester is examining how humans distinguish among billions of fellows and how much and what type of information we need to recognise a face.
Combining psychology and computer science the aim is to produce a generative computational model of human facial processing, to assimilate and comprehend the powers of the human brain.
The study’s conclusion could have applications to security, particularly entry-control, and in computer-generated characterisation in the film and TV industry.
Sequence is key
Dr Nick Costen, of MMU’s Department of Computing and Mathematics and principle investigator, said: "Seeing a face from a different angle, or with a different expression or simply seeing it fleetingly or in motion can affect recognition, but to what extent?
"We think recognition is significantly more accurate when we are exposed to realistic sequences of images. When faces are not familiar there is strong evidence that what matters is the sequence of movements."
This 3-year project is a collaboration between experts in computer vision techniques for recognising faces and experts in the psychology of how visual interpretation.
Images of 200 volunteers, both British and Japanese, will be presented in varying forms, (clear and distorted, still and moving) to a group of observers to test their levels of recognition.
Familiar or not?
Added Dr Costen: "Observing different types of faces in differing circumstances should give us a much rounder picture of the factors involved in human interpretation."
The computer side of the project will construct a generative computational model of human facial processing, able to cope with variations in familiarity. The model will be used to generate data for psychological experiments investigating the effects of movement on recognition.
Cognitive Systems Foresight: Human and computer face recognition from video sequences is an EPSRC/BBSRC funded project, valued at £413,000.
Dr Costen will work alongside two colleagues at the University of Manchester – Dr Tim Cootes, a reader in computer vision and Dr Karen Lander, lecturer in cognitive psychology.
For more abot computing research at MMU, go to www.docm.mmu.ac.uk/research/index.html