RIVR results show biometric liveness detection effectiveness highly variable

The state of the art in biometric presentation attack detection (PAD) is better than document validation, but far worse than matching. That much is clear from the Phase 3 results of the Remote Identity Validation Rally (RIVR). The test, conducted at the Maryland Test Facility (MdTF) for the U.S. Department of Homeland Security’s Science and Technology Directorate (S&T), shows wide variability in the performance of biometric liveness detection solutions on the market.

Phase 1 tested matching selfie biometrics to ID documents, and showed that most submissions could do the job, some even well. It also forecast the Phase 3 results. Phase 2 tested document validation, with soberingly poor results. The results from the PAD track are more encouraging, if not by much.

RIVR runs with predefined sets of performance goals and thresholds for several metrics. In the case of PAD, they are Bona Fide Presentation Classification Error Rate (BPCER), Attack Presentation Classification Error Rate (APCER), satisfaction and average transaction time for active and average run time for passive.

Six active and 12 passive PAD systems were assessed, their names anonymized in the results.

Of the six active systems, two met the threshold for BPCER, two met the threshold for APCER and another met the goal. Satisfaction rates were good for five of the six, with the other, PAD-A 6, delivering the worst results in each category.

Among the dozen passive systems, three met the BPCER goal and five more met the threshold, but only one (PAD-P 9) met the APCER threshold.  Only two of the other 11 were even close. All met the average run time threshold. Taken together, the results indicate PAD-P 1 is the only other system in this set that could be considered close to effective for the kind of scenario RIVR is designed to test.

Paravision supplied the PAD-P 9 entry, which was the only one to meet a RIVR goal in one category and the threshold in all others. PAD-P 9 had a BPCER below 0.5 percent, an APCER of 1.7 percent, and was around the middle of the pack in runtime, at 2.3 seconds.

“These results highlight a meaningful shift in the market. Passive liveness has often been viewed as the weaker option compared to active approaches that require user interaction, but that is simply no longer the case,” says Paravision CPO Joey Pritikin. “Our results demonstrate that with a sophisticated, AI-powered approach, passive PAD can deliver both robust fraud protection and a dramatically better user experience.”

Idemia Public Security announced that it also met all performance thresholds, which means it is PAD-A 5. IPS offers both active and passive PAD. PAD-A 5 had a 4.8 percent BPCER, a 3.3 percent APCER, 93 percent satisfaction and a 23 second average transaction time.

Notably, PAD-A 2 missed the 5 percent BPCER threshold by 1 percent, but otherwise met all goals, including with a 0 percent APCER.

Article Topics

biometric liveness detection  |  biometric testing  |  biometrics  |  DHS S&T  |  Idemia Public Security  |  Maryland Test Facility (MdTF)  |  presentation attack detection  |  Remote Identity Validation Rally (RIVR)