Iris biometric verification, palm prints tests and cheat sheets coming from NIST

The U.S. National Institute of Standards and Technology has plans to expand its biometrics evaluations and reports into several areas that are emerging as priorities with advances in technology and emerging applications.

Those plans for the future were previewed by NIST’s Patrick Grother in a webinar hosted by the Security Industry Association. “The Future of NIST Technical Evaluations of Biometric Technologies” was sponsored by Leidos and SAIC.

Grother explained the types of evaluations that NIST performs for face biometrics and related technologies, and the proprietary databases it uses. He also reviewed how they are used by developers, governments and businesses, to improve the technology in the former case and to ensure it works well in the latter two.

He also briefly touched on NIST’s work with iris, fingerprint and voice biometrics, as well as tattoo recognition.

‘Is it AI?’

Grother is often asked these days if these technologies, as well as facial recognition, are “AI.” The answer matters due to the trend towards regulators, particularly in Europe, imposing restrictions on biometrics use via legislation on AI.

“I asked the European regulators: ‘Would you seek to regulate an algorithm that’s demonstrably not AI?’ and they said ‘Yes,’” Grother recounts.

He notes that the original iris recognition algorithms were “not even machine learning,” but were hand-crafted by Cambridge Professor John Daugman.

Future work by NIST on iris biometrics, in addition to the IREX X 1:N evaluations, includes the launch of a new IREX XI to evaluate 1:1 comparisons. The agency will also double the size of its database for the 1:N test, and add a track for “difficult” images. The 1:1 evaluation will provide a lower barrier to entry for academics, and an appropriate forum for exploring challenges like noise or images captured with lower-resolution cameras.

NIST is also working on a metric to evaluate the speed and accuracy of iris algorithms together, as iris recognition has slowed down significantly over the years, relative to other biometric modalities.

Friction ridges, tattoos and filling in gaps

For fingerprints, NIST is relaunching a Friction Ridge (image + features) Technology Evaluation to look at distal fingerprints and palms.

NIST is also planning to re-start tattoo recognition work discontinued around a decade ago in 2025. It will look at 1:N recognition, as well as detection and image quality.

Grother says that NIST has considered evaluations for biometric template protection and privacy preservation through de-identification. NIST would like to stand up new datasets for selfies and driver’s license scans, but does not have access to the right kind of data. NIST has also thought about evaluations for sex and ethnicity classification, which may be of interest to European regulators, but has no plans to run such tests at this time.

NIST testing of presentation attack detection is on hold, pending the acquisition of more data.

Age assurance evaluations will continue, and may add assessment of confidence values for estimations of whether an age assertion, such as “I am over 21 years old,” is true. The agency will also look at noise in age estimation caused by “nuisance factors” like pose and movement in videos.

Grother emphasized the improvement in facial recognition accuracy over recent years, in particular due to tolerance for lower quality images. NIST wants to develop quality tests, therefore, that can find extremes in image quality factors that can make an image impractical for biometric matching.

Plans also include extending the duration of time between probe and reference images used in facial recognition from 10 and 12 years in current tests to 20 years.

Reports are also being prepared on traceability, demographics in 1:N matching, morph detection deployment and face risk management. NIST is also working on “cheat sheets” which provide guidance on reading its biometric test results and reports. The traceability report will address the gap between the prototype algorithms evaluated for the FRTE and those deployed in the field.

Among the ongoing evaluations, NIST intends to add an extension to its one-to-many face biometrics leaderboard to summarize demographic differentials.

As applications like digital travel credentials (DTCs) pick up steam, NIST also plans to issue a report on face biometrics for people on the move in the first quarter of next year.

Article Topics

biometric testing  |  biometrics  |  biometrics research  |  facial recognition  |  fingerprint biometrics  |  iris biometrics  |  Leidos  |  NIST  |  palm biometrics  |  SAIC  |  Security Industry Association (SIA)