NIST researchers used 89 different facial recognition systems, including those employed by the Department of Homeland Security and U.S. Customs and Border Protection, to conduct the study. All of the algorithms had an error rate of about 0.3 percent. However, when presented with subjects wearing face masks, those error rates increased from anywhere between 5 percent to 50 percent.
The team also found that the style of the mask and how it was worn affected the algorithms’ accuracy. For example, a mask that covered the wearer’s nose made them more difficult to identify than when the mask was worn below their nose. Additionally, the algorithms were more accurate when faced with round masks like N95s that cover less of the face as compared to masks with more coverage.
“None of these algorithms were designed to handle face masks,” Mei Ngan, a computer scientist with NIST, said. “With respect to accuracy with face masks, we expect the technology to continue to improve.”
