The Department of Homeland Security (DHS) found that facial recognition software used to identify travelers is more than 99 percent accurate on average, according to a Jan. 17 report.
The report comes after DHS instituted a directive requiring thorough testing of facial capture and facial recognition software to reveal unintended bias in its technology.
“We’re not perfect, but this report highlights how [facial recognition] is delivering real value for the public and supporting critical law enforcement missions – while maintaining our commitment to transparency, accountability, and responsible AI use,” Eric Hysen, the former DHS chief information officer and chief AI officer, said in a LinkedIn post.
DHS tested eight methods, which are primarily powered by artificial intelligence and used by Customs and Border Protection, the Transportation Security Administration, and Homeland Security Investigations.
Across the eight methods, the report found differences in facial recognition capabilities between volunteer participants with lighter skin tones compared to participants with darker skin tones.
In testing CBP technologies, face matching succeeded 98 percent of the time for people with darker skin tones compared to more than 99 percent of the time for people with lighter skin tones.
There was also a difference in CBP global entry touchless portals with face-matching technologies across age groups yielding a 97 percent success rate for matching 18–30-year-olds and more than 99 percent success rates for other age demographics.
In testing TSA technologies, similar results were found across self-identified race descriptions. Credential Authentication Technology (CAT-2) facial matching succeeded more than 99 percent of the time for self-identified white volunteers and 98 percent of the time for self-identified Black volunteers.
According to its website, TSA has implemented CAT-2 technology in 84 airports nationwide with plans to expand to over 400 in the coming years.
In one prototype technology test for TSA Precheck, there were significant disruptions in facial capture technology. The technology is used to verify a face is present before taking a picture and showed success rates between 88 percent and 97 percent, noting less overall success for volunteers with darker skin tones.
DHS is implementing a manual capture option for this technology to minimize future disruptions.
“TSA and DHS are evaluating new algorithms to improve this step and plan to implement them later this year,” the report reads.
“Overall, FR/FC systems performed extremely well for diverse demographic groups,” it adds. “On average, the technology worked more than 99 percent of the time for systems that are fully operational.”
