RICHARDSON — With 56 million travelers coming through Dallas/Fort Worth International Airport each year, keeping tabs on security takes more than eyes; it takes machines that can recognize human faces.
Now, researchers at the University of Texas at Dallas are looking at ways to make machines better at recognizing human faces than humans are.
"It's one possible way to think about improving security in the future," said Alice O'Toole with UT-Dallas Brain and Behavioral Sciences.
Security companies are racing to create effective facial recognition software, or "algorithms."
"The algorithms are not going to get bored, and they're not going to have their attention diverted to something else going on," O'Toole explained.
Her research analyzes facial recognition algorithms to see if they're better than people. They both run through a highly controlled test that includes two faces on the screen — each with neutral expressions — each shot in a different location.
So, which is better: People or machines?
"On average now, the very best algorithms in the world, on the tasks that we looked at today, are better than humans," O’Toole concluded. "Pretty similar, but a little bit better."
Controlled settings are one thing, but the real world is full of variables. In fact, Arkansas, Indiana, Nevada and Virginia all outlaw smiling on driver's license photographs. A smile can trick the algorithms that root out fraud by comparing new photos with existing ones in a state's database.
"When you smile, it changes the relative positions of the eyes, the features, the nose may look a little different because the position of the mouth," O'Toole said. "You see dimples sometimes that weren't there."
In the end, O'Toole finds combining algorithms and people together dramatically improves the effectiveness of facial recognition. The machine quickly presents possible matches from a giant database; then the human can sort through the options to make a match.
Had that kind of system been available on September 11, 2001, O'Toole says it could have helped.
"If you have a good watch list, I think it's certainly a good start," she said.
E-mail dschechter@wfaa.com
