
The Sciences
60-Second Science
Subscribe:
Look At That Face: Improving Automatic Face Recognition
An average of 10 or more photos of a person is easier to match up with a face seen in real life than is any individual photograph of the person. Steve Mirsky reports. Get a free audiobook at www.audible.com/science
Face recognition software is increasingly important in areas ranging from surveillance to internet image searches. But research shows that face recognition systems do a poor job matching a face in a photograph to a face seen in real life. In fact, most people are pretty poor at it, too. Which is one reason why looking at mugshots can be unreliable. But researchers from the University of Glasgow have developed a new system that greatly improves the chances of matching a face with a photo, for both computers and us. They announced their findings this week at the Festival of Science in York, England.
You know how the occasional photo of you just doesn’t quite look like you? Well, the key to the new process is to use 10 or 11 different photos of a person and average those together to get a new image. Such averaging does away with odd effects in an individual shot that could be due to lighting or just the way you happened to be holding your face. Because your average face is more recognizably you, even though you’re well above average.