ROC impresses in NIST biometric age estimation benchmark, Shufti makes debut

Two new entrants to NIST’s Face Analysis Technology Evaluation (FATE) Age Estimation & Verification, one a debut and the other a return, illustrate the widening supply of biometric facial age estimation and age verification options for online businesses, and the steady incremental improvement in the technology’s accuracy.

ROC and Shufti submitted new algorithms for the February 13 update of the ongoing evaluation.

The roc_002 algorithm showed particularly impressive results, with the lowest mean absolute error (MAE) rate among all submissions for South Asian females (2.5 years, plus or minus 0.08) and both females (2.9 plus or minus 0.2) and males (2.5 plus or minus 0.2 ) from West Africa. The same algorithm also placed among the ten lowest MEAs as of the mid-February update for every other demographic except East Asians, for whom it had the 12th and 13th-lowest error rates.

The ROC algorithm delivered similar results in mixed datasets and the Challenge 25 and Child Online Safety (13-16) evaluations, though its FPR came 15th for ages 8-12 and 25th for ages 17-22.

Shufti’s first entry into the FATE age report showed some room for improvement, though the shufti_000 algorithm did crack the top 15 in metrics within the Challenge 25 and Child Online Safety categories.

Mean Absolute Error (MAE) by Interocular Distance was added to the results page with the previous FATE Age update in January. Regula, Cognitec, Yoti, Idemia and Innovatrics continue to hold spots near the top in all evaluated metrics.

Article Topics

age verification  |  biometric testing  |  biometrics  |  Face Analysis Technology Evaluation (FATE)  |  face biometrics  |  facial age estimation (FAE)  |  NIST  |  ROC  |  Shufti