Academic Minute
5:00 am
Tue June 3, 2014

Dr. Megan Papesh, Louisiana State University - Flaws of Facial Recognition Tech

Facial recognition technology is still in its infancy.

Dr. Megan Papesh, assistant professor of psychology at Louisiana State University, is demonstrating weakness in a system many people may assume is completely secure. 

Dr. Megan Papesh - Louisiana State University

Dr. Megan Papesh is an Assistant Professor of Psychology at LSU, and the director of the Eye Movements, Memory and Attention Laboratory. Her research on face matching has been featured in the Huffington Post, Wall Street Journal and on NPR. Her lab also researches topics related to episodic (and autobiographical) memory, language processing and visual search.

Papesh Eye Movements, Memory and Attention Lab

LSU Psych Department

More on Facial Recognition's Current Fallibility

LSU Research News

Megan Papesh - Flaws of Facial Recognition Tech

The fact that two people who boarded the lost Malaysian Airlines flight 370 used stolen passports raised a valid concern: how do people get through security with IDs featuring photos of other individuals? Because society relies on face recognition and ID verification for so many tasks, people are under the impression that we are experts in this domain. Our research shows the precise opposite.

One of our most recent studies finds face matching related to ID checks to be incredibly fallible, with error rates between 10 and 20 percent under ideal, laboratory-induced conditions. But in most high-risk situations, including passport control, where people can assume a very low rate of fake or stolen IDs due to the level of difficulty inherent in obtaining such false documentation, the numbers tell a different story. When observers infrequently encountered fakes, they failed to catch 45 percent of them, even when given multiple opportunities to correct their errors. That’s a tremendously high rate when you consider the implications for terrorism and criminal activities.

When we encounter someone we’ve seen before, even if their appearance has changed, our brains make compensations allowing recognition. For unfamiliar faces, we must extract information from both a photo ID and a human being, then undergo a perceptual search using nonspecific, imprecise targets.

Relying on instantaneous identity verification using a photo that captures us frozen in a single moment of time is setting ourselves up for extreme vulnerability. Ours is not the only study demonstrating weaknesses inherent in face matching. In the end, a better understanding of face-matching might lead us to a better way to effectively use photo IDs for their intended purpose.

Related Program