RIVR results show targeted selfie biometric spoof attacks divide the field

Test results from the U.S. Department of Homeland Security’s Science and Technology Directorate (S&T) suggest significant differences in the performance of selfie biometrics systems between established providers.

Phase 1 of the Remote Identity Validation Rally (RIVR), carried out at the Maryland Test Facility (MdTF), focussed on matching selfie biometrics to ID documents.

In all, 10 of the 16 systems submitted by biometrics vendors for the first phase of RIVR met all of the performance “thresholds” set for RIVR, but only 5 met the more stringent “goals.”

Performance in the Selfie Match to Document (SMTD) track began breaking down for some technologies as early in the process as properly cropping and extracting the facial image from the ID, MdTF Researcher and SAIC Scientist Director Yevgeniy Sirotin said in a webinar on the Selfie Match to Document Results. The actual selfie collection and processing was not a challenge for participating systems. Failures to extract images from documents were associated with demographic differences for a couple of vendors, but the majority were consistently low for all demographics.

A few systems struggled with false non-match rates (FNMR) on all three smartphones used, which Sirotin said may be related to configuration or challenges in processing the images. Three vendors had FNMRs above 50 percent.

In false match rate (FMR), S&T found demographically similar imposters resulted in an 11-times higher error rate on the median system. Only seven of 16 systems delivered acceptable FMRs for demographically similar spoof attempts.

Shufti and Paravision among leaders

Shufti and Paravision are two of just five among the 16 participating vendors to meet all the performance goals set by S&T.

Shufti states in a post on the results that it had a FMR below 0.01 percent and a FNMR at 1 in 10,000 below 0.68 percent, which identifies Shufti’s entry as MTDS 15.

For demographically similar comparisons, Shufti says it had 50 false matches out of 250,000 attempts, and that this demonstrates “strong impostor resistance even under more realistic attack scenarios.”

Paravision reports it was MTDS 12, with an FMR below 0.01 percent and an FMNR below 0.08 percent.

The company says it was the only vendor to meet RIVR’s goal of an FMR below 0.01 percent for demographically similar imposters while also meeting the goals in all other metrics.

“These results show that strong remote identity verification isn’t just about matching faces – it’s about resisting intentional, targeted fraud under real-world conditions,” says Paravision CPO Joey Pritikin. “We’re proud to see Paravision recognized by DHS S&T for delivering both accuracy and security where it matters most.”

Participants aliased as MTDS 1, 4 and 10 also met the FMR and FNMR goals of the Rally.

RIVR’s second track focusses on document authenticity, and DHS invited participants for the third, addressing biometric spoof protection, last August.

Article Topics

biometric matching  |  biometric testing  |  DHS S&T  |  Maryland Test Facility (MdTF)  |  Paravision  |  remote identity proofing  |  Remote Identity Validation Rally (RIVR)  |  selfie biometrics  |  Shufti  |  spoof detection