Biometric age estimation graduates with high marks from NIST, ACCS

The inaugural evaluation of biometric age estimation technologies previewed at the Global Age Assurance Standards Summit 2024 has been published. It shows significant differences in accuracy among the half-dozen developers submitting algorithms.

U.S. National Institute of Standards and Technology Biometric Standards and Testing Lead Patrick Grother told a Summit audience during the preview that the results show that biometric age estimation is potentially as effective as checking a date of birth field on an ID document.

The newest track in NIST’s Face Analysis Technology Evaluation (FATE) activity breaks down age estimation algorithms according to their effectiveness for “Challenge 25” scenarios, like determining if a young adult is old enough to buy age-restricted goods, “Child Online Safety” scenarios, which focus on ages 13 to 16, and resource usage.

Yoti, ROC, Dermalog, Incode, Neurotechnology and Unissey are the first crop of age estimation developers to submit algorithms to NIST for evaluation.

The report shows false positive rates (FPRs) for the Challenge 25 scenario ranged from Incode’s 0.047 (±0.017) to Neurotechnology’s 0.353 (±0.036) in the application category, and Unissey’s 0.130 (±0.025) to Dermalog’s 0.330 (±0.036) in the border category. For child online safety, mean absolute error rate ranged from 2.495 (Yoti) to 5.188 (Neurotechnology). FPRs ranged from 0.055 for under-13s from Incode and 0 for 17-and-overs from ROC to 0.231 and 0.009, respectively, both from Unissey.

Incode says its results not only demonstrate the lowest mean absolute error (MAE) rates for multiple datasets, but also demographic fairness and tolerance of real-world conditions.

The report also shows dramatic improvements in the state of the art since 2014, and explains the origin of NIST’s datasets.

More biometrics developers are expected to submit algorithms for future evaluations, and the publication of the report coincides with market moves by a couple of likely candidates (see below).

Age assurance summit findings

The Global Age Assurance Standards Summit has published documentation of its own, in the form of a communique on the main take-aways from the ground-breaking conference last month in Manchester and a compendium of the presentations and insights shared during the event.

The communique, published by the Age Check Certification Scheme, emphasizes that age assurance is realistic, effective, and ready to one tool in a range of measures for protecting children online. International standards will help, and laws and regulation can provide the necessary legal framework and enforcement procedures.

“If deployed proportionately and effectively, Age Assurance represents an opportunity to enhance the fundamental rights of children in a digital age, in addition to protecting anonymity and the freedoms of adults to enjoy online goods, content and services,” the communique reads.

The communique sets out the mandate of policy-makers to address the risks of uncontrolled access by children to online goods, services and content, and the international standards that can provide the basis for doing so. It also reiterates the call to action and principles outlined at the culmination of the event.

The 88-page compendium includes synopses of more than 200 sessions from the week-long summit, dozens of hours of video and dozens more slide decks. There were over 200 attendees at the event, and another 424 online (including Biometric Update).

Market options increasing

Trust Stamp has filed for patent protection of its method for improving the performance of biometric age estimation algorithms with AI.

The company says its new algorithm allows it to reduce error rates and more easily tailor the results provided by its biometric age estimation product unveiled in March to meet the complex requirements of clients.

“The algorithm builds on the Bayesian framework to not only address the inherent uncertainty in predicting human’s age but also in ways that can reduce the effect of over and under estimation had raw age values been used,” says Trust Stamp CSO Dr. Norman Poh. “Moreover, the solution allows us to produce age category probabilities that can be easily tailored to different clients’ requirements and even to target demographics, thus making the age estimation much more reliable and targeted.”

Privately SA and Luciditi, developed by Arissian, have announced a strategic partnership to combine their age verification technologies into a comprehensive solution for the retail, gaming, gambling, social media and ecommerce industries.

The new partners plan to combine Privately’s on-device age estimation with the ID document scanning and digital credential issuance capabilities of Luciditi’s platform. Both technologies protect compliance and user privacy, according to the announcement.

The partners hold a bundle of industry certifications related to age assurance and privacy protection.

Article Topics

Age Check Certification Scheme (ACCS)  |  age verification  |  biometric testing  |  Face Analysis Technology Evaluation (FATE)  |  face biometrics  |  Luciditi  |  NIST  |  patents  |  Privately  |  standards  |  Trust Stamp