iBeta biometrics lab busy with 5 PAD tests, 1 for accuracy in last 2 months

Half a dozen biometrics providers have received letters from iBeta Quality Assurance confirming their compliance to international standards for presentation attack detection or biometric accuracy within the past two months.

Five of the six tests used the ISO/IEC 30107-3 standard, and tested for defense against Level 2 attacks.

Advance.AI is the latest to receive compliance confirmation for testing to Level 2 of 30107. The company’s Liveness Detection software allowed no successful presentation attacks in testing with a Samsung Galaxy S22 and an Apple iPhone 12.

The Youverse SDK was tested with a Samsung Galaxy A32, and iBeta was unable to spoof a liveness classification with the presentation attacks.

Brazilian ID verification and background check provider idwall’s Face Liveness application did not return a liveness classification with presentation attacks on a Galaxy S22 and an iPhone 12.

Tests of Brazil-based Payface’s Fortface selfie liveness software likewise did not return any liveness classifications in presentation attacks using a Galaxy S20+ and an iPhone 13, the confirmation letter states.

Veriff’s Biometric Liveness software was tested on an iPhone 15, with a 0 percent imposter attack presentation accept rate (IAPAR).

Bona fide presentation classification error rate (BPCER) or bona fide false non-match rate (FNMR) is found in the final report for each, which are not publicly released.

VNPay passes accuracy test

The 30107 standard applies to liveness detection, rather than accuracy, and specifies a BPCER or FNMR limit of 15 percent. The ISO/IEC 19795-2 standard is for accuracy, not presentation attack detection, and specifies a false rejection rate (FRR) of 5 percent or lower.

VNPay’s liveness detection was tested against the 19795 standard. The VNPay Face EKYC v1.0 software was tested on an iPhone 13 Pro Max. The testing showed a 0.001 percent false acceptance rate (FAR) at a confidence threshold of 95 percent, and no false accepts were found during the test. The false reject rate (FRR) at a 95 percent confidence interval was found to be 0.0024 percent FRR from bootstrapping. False match rate (FMR) and false non-match rate (FNMR) were both found to by 0 percent.

Failure to acquire and failure to enroll rates are found in the final report.

VNPay also passed a Level 1 PAD test from iBeta last November.

Article Topics

accuracy  |  Advance.AI  |  biometric liveness detection  |  biometric testing  |  biometrics  |  iBeta  |  idwall  |  ISO/IEC 19795-2  |  ISO/IEC 30107-3  |  Payface  |  presentation attack detection  |  Veriff  |  VNPay  |  Youverse