NIST sees biometrics developers closing in on operational morph detection

Biometrics developers are getting better at detecting face morphs, even if facial recognition algorithms aren’t. Day 2 of the EAB’s workshop on “Face Morphing: State of the Art & Outlook,” co-organized with Annalisa Franco of the University of Bologna, Italy, offered a measure of optimism to balance the startling threat that AI-blended photos pose to global border security systems.

Presentations on the first day of the workshop focused on developments in morph creation, metrics for vulnerability to morphing and the potential of different kinds of morph attack detection (MAD) for use at different points in the identity lifecycle.

Mei Ngan of NIST gave an update on the agency’s Face Analysis Technology Evaluation (FATE) MORPH track. It shows that the most accurate facial recognition algorithms today are still highly susceptible to face morphs.

Ngan showed a set of graphs with leading developers Cognitec, Idemia, Innovatrics, Neurotechnology and ROC which showed that while their overall miss rates have declined markedly over the past seven years, their vulnerability to morphs is improving little if at all.

NIST adds new morphs to its dataset for MAD evaluation as new techniques emerge, Ngan says, including recent additions of two diffusion-based methods. The results are compared with those matching bona fide images.

The evaluation includes S-MAD and D-MAD systems, as well as morph-resistant 1:1 face biometrics algorithms. Another sub-track evaluates demorphing.

Among NIST’s overall findings is that generalization to novel morph species remains a challenge in S-MAD.

MAD may by ready for background pilots

The best MAD systems are very close to delivering operational value, Ngan argued, using one of the leading D-MAD algorithms in the FATE MORPH evaluation as an example. It would currently return one false positive in a hundred while catching roughly 72 percent of biometric face morph attacks.

There could be value in deploying them “behind the scenes” and investigating retrospectively to see if flagged cases were morphs, Ngan suggests.

NIST published a document last August to help organizations prepare for implementing MAD in anticipation of those next steps.

Ngan delved into different operational and investigative methods for detecting face morphs, noting for that the use of multimodal biometric verification, like at Singapore’s Changi Airport, could be effective. Initial indications are that in passport renewal scenarios, one-to-many facial recognition could be used to flag morphs, based on finding high scores for multiple subjects.

NIST is also hopeful that if developers focus on “the twins problem,” then “the morphing problem may naturally go away.”

In the long run, more trusted biometrics capture processes should be part of the solution, Ngan argues.

Day 2 of the workshop also featured presentations on Horizon Europe research projects on face morphing.

Article Topics

biometric testing  |  EAB  |  EAB 2026  |  Face Analysis Technology Evaluation (FATE)  |  face biometrics  |  face morphing  |  morphing attack  |  Morphing Attack Detection (MAD)  |  NIST