Yoti shows further facial age estimation accuracy gains ahead of OSA deadline

Yoti’s facial age estimation technology has hit a new level of accuracy in internal testing of the approved method for “highly-effective age assurance.”

The accuracy figures are revealed in a new white paper published days ahead of the UK’s Online Safety Act taking effect, and forcing digital platforms to institute age checks using a “highly effective” method like facial age estimation. The new responsibilities under the OSA kick in on July 25.

The mean error rate (MEA) for Yoti’s technology estimating the ages of people between 13 and 17 years old is now 1.1 years. This is particularly important for the 15- and 16-year-old age limits for social media. For 18-to-24 year olds, Yoti’s MAE is now 2.1 years.

The testing also shows that Yoti’s technology will correctly estimate 99.3 percent of youths between 13 and 17 as being under 21 (as measured by true positive rate, or TPR), and 99 percent of children between 6 and 12 as being under 13. The variance based on skin tone for TPRs of 13-17 year-olds across skin tones 1, 2 and 3 was 0.7 percent, which the company describes as “no discernable bias.”

MEA for people 6 to 70 years old is 2.4 years.

“With a growing demand for effective age assurance and increasing regulatory engagement, we have continued to invest and improve the accuracy of facial age estimation,” says Yoti CEO Robin Tombs. “It is fast becoming an essential tool to help protect young people online, create safer, age-appropriate experiences and respect people’s privacy. The improved accuracy means even more people will now be able to use our privacy-preserving facial age estimation to easily prove their age online.”

The white paper also provides guidance on safety thresholds, buffers and “the waterfall approach,” explains evaluations of facial age estimation technology by NIST and ACCS, and offers lessons from experience.

Online age assurance checks in the UK are about to spike, with platforms required to verify whether their users are children or not if they host any adult material. That means pornography platforms, but also peer-to-peer sites and search engines, instant messaging services, dating apps and others are obligated to use age verification or estimation.

“The majority of adults choose our facial age estimation to quickly and easily prove their age. For those estimated below the threshold, they can use our privacy-preserving Digital ID app or share their age from an official ID document,” Tombs told Biometric Update in an emailed statement about the OSA coming into force on Friday. “As age checks become more common, nearly 1 in 4 people are using a reusable age token or Digital ID to anonymously prove their age. We’re proud to be protecting children online and introducing people to privacy-first age solutions – because online safety shouldn’t come at the cost of privacy.”

Article Topics

accuracy  |  age verification  |  biometric age estimation  |  biometric testing  |  biometrics  |  face biometrics  |  facial age estimation (FAE)