Systems designed to identify criminals and suspects from photographs are often unreliable due to image variability such as lighting and pose.
However, new findings by University of Glasgow researchers, and reported in the journal Science in January, show that by combining a number of photographs into an average image of the person, the face recognition results are twice as accurate.
Lead researcher Dr Rob Jenkins said: “We were surprised by the level of success of our trials. A photo of a face captures a single moment and two pictures of the same person can look quite different depending on when they were taken, as well as changes in lighting or pose. With image averaging, we can wash away all these differences, and extract the true essence of the person’s face.
“The applications of this technology are quite extensive, from the identification of missing persons to use in customs. In fact the face recognition technology, FaceVACS, is being tested in Sydney airport." The research was developed by studying how the human mind recognises faces.
Co-researcher Professor Mike Burton said: “As humans we are amazingly good at recognising people we know, but we are actually very bad at matching someone we don’t know to their photo. In this project we have borrowed from psychological research on how we recognise familiar people, and discovered that this can substantially improve automatic face recognition too. This may offer a way forward for technology that has not yet lived up to its promise.”
The researchers used the face recognition technology FaceVACS and the website MyHeritage.com, containing over 30,000 photos of celebrities, to test their findings.
