NEC facial recognition takes top score in half of latest NIST 1:N test categories
NEC has claimed the highest biometric accuracy score in four of eight categories in the latest assessment from the U.S. National Institute of Standards and Technology’s 1:N Identification track, as of the January 22, 2024 report.
NEC scored the top accuracy assessment for matching mugshots against a database of 12 million, at 99.88 percent. The company’s biometric algorithm also topped the Kiosk, Border at ten or more years and Mugshot at 12 or more years categories. It came second in three categories, and third in the remaining one (Mugshot N=16 million).
The company’s facial recognition business operates in more than 50 different countries and regions, according to an announcement, and is used at roughly 80 airports. NEC’s NeoFace Monitor is also used for 1:1 facial authentication by more than a thousand organizations.
The FRTE 1:N compares false negative identification rate (FNIR) with the false positive identification rate (FPIR) set to 0.003 (three in one thousand).
How accurate is accurate enough?
NIST Applied Cybersecurity Division Digital Identity Program Lead Ryan Galluzzo said during a recent event that his team is working on how to integrate testing and evaluation into guidelines for AI and digital identity, Federal News Network reports.
Galluzzo was speaking during the 2024 Policy Forum hosted by the Better Identity Coalition, FIDO Alliance, and the Identity Theft Resource Center. Operational testing and continuous monitoring of how the technology is working are also key, he says.
Standards and testing for biometrics are further ahead than many of the other applications his group deals with, but he emphasizes the importance of “getting the right representative data to make sure you’re actually representing your population.”
For its part, the American Civil Liberties Union says there is no “magic” number for accuracy at which facial recognition is safe. Considering the true positive rate alone is not sufficient to fully understand an algorithm’s effectiveness.