ID R&D claims leadership among ‘extremely accurate’ leaders in NIST PAD test
ID R&D is claiming a top spot for the security and convenience of its printed photo and screen replay attacks in biometrics testing by the U.S. National Institute of Standards and Technology.
NIST tested 82 passive biometric presentation attack detection algorithms for its first Face Analysis Technology Evaluation (FATE) Part 10 for PAD algorithms.
The facial liveness algorithm submitted by ID R&D achieved an attack presentation classification error rate (APCER) near zero for photo print and replay attacks. This is critical to secure remote onboarding processes against fraud without making the process inconvenient for users with false positives, according to the company, due to the commonality of these attack types.
Alexey Khitrov, CEO of ID R&D, told Biometric Update in an email that the results make clear that for these common presentation attacks, “passive PAD can be extremely accurate.” Beyond the top performers, he cautions, “performance seems to fall off pretty quickly.”
The test evaluated different use cases, and Khitrov notes that the “evasion” use case is not applicable to developers working on security for remote onboarding.
Evasion is applicable to forensic and surveillance scenarios, in which users are attempting to find a certain individual. The company has also posted an FAQ to explain NIST’s evaluation and its results. It delves further into the difference between the different types of attacks tested.
“We were also the overall top performer, with the highest number of #1 rankings and lower error rates across all the impersonation tests,” he says. “People sometimes can’t believe that passive liveness could work so well. Hopefully these results demonstrate that in fact a passive approach is extremely effective, with the benefit of a frictionless UX that doesn’t tip off fraudsters.”
“NIST is among the world’s most respected experts in assessing commercial biometric algorithms, and so we are appreciative of their work on PAD and thrilled with the results of this thorough evaluation,” says Khitrov in the announcement. “For our team, our performance serves not only as validation of our belief that biometric security should be frictionless, but also compelling evidence that the current generation of liveness is capable of delivering both security and convenience.”
“Using just a single image frame, we’ve managed to achieve near-zero error rates with a solution that is completely transparent to users, setting a new benchmark for the industry,” says Chief Security Officer Konstantin Simonchik.
In the future, Khitrov says he would be interested in learning from NIST’s expertise in facial injection attacks, document presentation attacks, and document injection attacks.
Article Topics
biometric liveness detection | biometric testing | biometrics | Face Analysis Technology Evaluation (FATE) | ID R&D | NIST | presentation attack detection
Comments
One Reply to “ID R&D claims leadership among ‘extremely accurate’ leaders in NIST PAD test”
Leave a Reply
This site uses Akismet to reduce spam. Learn how your comment data is processed.
All of these “2D Liveness” vendors should be ashamed of their performances on this NIST PAD test. Most understand that and have kept quiet. Only ID R&D dares to insult your intelligence.
Don’t believe what Russians say about security; .6% is not near 0…
ID R&D only caught 99.4% in the PA Type 8 (zoomed). It’s one of the easiest spoofs in the real world to catch and should have been 100% caught.
I don’t know how these vendors can stay in business. It must be because they do tiny volumes because at our volumes, if ≥.6% of attacks were successful, that would mean allowing 20,000-50,000 attacks through a day…
These vendors do not have spoof bounty programs because they are not secure and would go broke from payouts if they did have them. The UK Home Office PAD testing proved it, and this NIST PAD Testing proved it again, especially for ID R&D, because they don’t have any active or randomness layers to increase security in their real-world deployments. In security, less is not more. Less is less.
Liveness Accuracy is logarithmic, the inverse of exponential hockey stick curves you normally see in tech. So, it’s quick and easy to get to 90% accuracy, 20x longer to get to 99%, and 10,000x longer to get to 99.9%. Let alone +99.9995% like FaceTec.
2D Liveness has gotten all possible easy gains; it won’t get much better in the next 20 years and, in fact, may very well get worse because of overfitting in the artifact training sets. This is why FaceTec uses 3D face data and why this NIST test is irrelevant except for showing that 2D Liveness & vendors that bet the farm on 2D will never be good enough for real security.