Identical twins really aren’t identical, as their parents or anyone who spends much time with them can tell you. The overall face may look the same, but the corners of the eyes may be slightly different, or the tip of the nose or the shape of the jaw.
“It’s these fine-scale differences that, in principle, may be more useful for distinguishing between lookalike twins than gross shape features,” says Patrick Flynn, a Notre Dame professor of computer science and engineering who studies biometrics.
Flynn and his colleague, Professor Kevin Bowyer, have been studying twins to learn how to sharpen the precision of face recognition software. “The thinking is if we can tease apart identical twin siblings through software, perhaps similar techniques can be employed to do a better job in distinguishing individuals in the population at large.”
Since 9/11 the federal government has been increasingly interested in the technology to enhance security. Flynn and Bowyer’s research on recognizing twins is funded by grants from the FBI and the U.S. Department of Justice.
The Notre Dame researchers have found that the more time people have to examine faces, the more likely they are to identify differences. “What is it that people are seeing that allows them to do that? That’s what we want to know,” says Bowyer.
The researchers are making plans to use eye movement tracking equipment to determine what facial features are receiving attention from the subjects while they are performing the identification task. Such features could be exploited by future generations of software to improve automatic decisions.
Current state-of-the-art face recognition software has a difficult time telling twins apart. People, however, are surprisingly adept at visually finding the subtle features that distinguish one individual from another. “It’s a matter of what features the software takes into account in judging whether a picture taken today and a picture taken previously, perhaps even years ago, represent one and the same person,” says Bowyer.