A facial recognition system similar to the one used by the U.S. Department of Homeland Security (DHS) had a 96 percent biometric accuracy rate with masked faces in testing as part of the 2020 Biometric Technology Rally held by DHS’s Science and Technology Directorate (S&T).
The Biometric Technology Rally was held at the DHS-affiliated Maryland Test Facility (MdTF), and tested 6 facial image acquisition systems and 13 biometric matching systems with 582 volunteers representing 60 countries, both with and without masks on. One iris recognition system was also tested. Subjects were identified one at a time in an environment roughly simulating airport conditions.
The median system performance was a successful identification rate of approximately 77 percent, but the best was successful for 96 percent of subjects.
The announcement by DHS S&T suggests that the results show organizations may be able to allow people performing biometric photo ID checks to keep their masks on, reducing the risk of spreading COVID-19.
“This isn’t a perfect 100 percent solution,” comments Arun Vemury, director of S&T’s Biometric and Identity Technology Center, “but it may reduce risks for many travelers, as well as the frontline staff working in airports, who no longer have to ask all travelers to remove masks.”
The worst system matched masked faces only 4 percent of the time, and the worst performance for people without masks was only successful at 11 percent of identifications. The camera failed to acquire an image in 6 percent of attempts on unmasked faces, and 14 percent for masked faces. For masked faces, the algorithm was unable to find the face in an additional 1 percent of cases. The iris recognition system collected an image only 80 percent of the time for people not wearing masks, and two-thirds of the time for masked faces, though it was also the fastest acquisition system.
Biometric acquisition system results are available from MdTF, and detailed matching system results are pending.
The companies are given aliases in the results shared by MdTF.
NIST testing shows facial recognition algorithms have improved at matching masked faces, though they are still significantly less effective than with unmasked faces.