The National Institutes of Science and Technology (NIST) released a report today that reveals significant failing with facial recognition technologies during the coronavirus pandemic. The U.S. Depart of Commerce agency’s study found that when faces are even partially obstructed by masks, error rates in detection spiked to between 5% and 50%. This compares to just 0.3% on average when no mask was worn.
While there is no federal mask mandate in the USA, some counties and states do require residents to wear masks. In addition, many national retailers require shoppers to wear masks prior to entry.
The facial recognition technologies study used photos with digitally applied masks to test 89 commercial facial recognition algorithms from Canon, Tencent, and Panasonic among others. The original photos served as a baseline.
Report coauthor and NIST computer scientist Mei Ngan noted, “With the arrival of the pandemic, we need to understand how face recognition technology deals with masked faces. We have begun by focusing on how an algorithm developed before the pandemic might be affected by subjects wearing face masks. Later this summer, we plan to test the accuracy of algorithms that were intentionally developed with masked faces in mind.”
NIST researchers used a variety of face shapes as well as nine styles of masks to reflect shape, coverage, style, and color differences. The researchers also used their AI algorithms on more than 6 million photos, which were previously part of facial recognition vendor tests.
The results of the NIST study won’t come as a surprise to anyone who owns an iPhone 11, X, XS, XR, or late model iPad Pro. Users of these Face ID-powered devices have experienced difficulty unlocking their devices while sporting masks or other facial coverings. Apple recently released an update to its operating system that immediately prompts users for their passcode if it detects the user is wearing a mask.
More information about the NIST report can be found at www.nist.gov.
