Yoti previews facial age estimation accuracy gains from new model

Age estimation based on face biometrics is getting steadily more accurate, and Yoti is reporting further gains from its latest model.

Robin Tombs shared a preview of the results it will detail in its next white paper on facial age estimation on August 24 in a LinkedIn post.

The accuracy of Yoti’s Facial age estimation for 18-year-olds has improved by a 7.8 percent reduction in mean absolute error (MAE), from 1.35 in the model introduced on March 23 to 1.25, according to the company’s internal evaluations. For 15 to 17-year-olds, MAE dropped from 1.09 to 1.02, and for 10 to 12-year-olds, MAE has dropped from 1.16 to 1.11.

For people between 6 and 70 years old, Yoti’s MAE is 2.56, down 11.3 percent from 2.88.

Tombs also reiterated criticism of comments from the CEO of communications regulator Ofcom, who suggested during a television appearance that age estimation “does not work very well for children.”

“Yoti customers, including some of the largest digital brands across several sectors, would not be paying to use our FAE at scale if the service did not work effectively,” he says.

Ultimately, Yoti is continuing to invest in facial age estimation to ensure it is effective for all types of age checks, whether subjects are above or below 18, 13, or any other age threshold set by regulators.

Tombs says Yoti will seek independent certification for its new model from the Age Check Certification Scheme (ACCS) and submit the model to NIST for future evaluation.

The dialogue between biometrics providers, lobbyists, advocates and policy-makers remains contentious, but the more time passes, the less arguments against using age estimation based on accuracy hold water.

Article Topics

accuracy  |  age estimation  |  algorithms  |  biometric testing  |  biometrics  |  face biometrics  |  facial analysis  |  Yoti