Age estimation leaders emerge in NIST evaluation

The National Institute for Standards and Technology (NIST) has released its latest Face Analysis Technology Evaluation for Age Estimation & Verification (FATE AEV) – “an ongoing evaluation of software algorithms that inspect photos of a face to produce an age estimate,” which measures the accuracy and computational efficiency of algorithms.

In NIST’s table displaying mean absolute error (MAE) values averaging 26 MAE estimates from two sexes and 13 ages, Idemia’s age assurance algorithm is on top, with MAE values under 3.0 for application photos (2.759), border crossing photos (2.746) and mugshots (2.361). Incode, Nominder, Jumio and Yoti round out the top 5.

Idemia is likewise on top of the evaluation by demographic group for men and women born in six regions of the world, demonstrating “low variation across columns,” which indicates “equitable accuracy across demographic groups.” Algorithms from Neurotechnology, Nominder, Yoti and Incode also appear in the top 5. Both Neurotechnology and Yoti have two algorithms in the top 10.

And in NIST’s Challenge 25 category, it is once again Idemia with the best results, posting an application False Positive Rate (FPR) of 0.024±0.011. Incode, Yoti, Nominder and ROC are the rest of the top 5, with a legacy Yoti algorithm placing sixth.

Despite the name of the evaluation, the test includes only facial age estimation algorithms using biometric data, rather than verifications based on comparisons to identity documents or other authoritative sources.

NIST’s first assessment of age estimation under the FATE program was published in May, and came as a dramatic expansion of the independently tested, publicly available data about the accuracy of inferring age based on face biometrics.

Article Topics

age verification  |  biometric testing  |  Face Analysis Technology Evaluation (FATE)  |  face biometrics  |  IDEMIA  |  Incode  |  Jumio  |  Neurotechnology  |  NIST  |  Nominder  |  ROC  |  Yoti