Bias testing introduced for biometric liveness detection in ID R&D, BixeLab collaboration
A new test protocol to evaluate the demographic differentials, or bias, in biometric liveness detection systems has been developed by BixeLab in cooperation with ID R&D, which commissioned the pioneering test.
ID R&D President and Founder Alexey Khitrov was joined by BixeLab Managing Director Ted Dunstone for a presentation and discussion on the novel evaluation of bias in liveness detection, in a webinar titled ‘Demographic Bias in Facial Liveness Detection: Impact, Measurement, and Mitigation’ and moderated by Biometric Update.
Khitrov explained the impact of bias in liveness systems, which like increased error rates in biometric matching algorithms, can result in unacceptable error rates among certain groups, often including those people with darker skin.
Dunstone outlined the test methodology and how it was developed, as well as why it focusses on BPCER (bona-fide presentation classification error rate) rather than APCER (attack presentation classification error rate).
BixeLab’s report on the process and an ID R&D white paper on the subject were also briefly summarized.
Both called for other vendors to address the challenge of disparities in performance among different demographics throughout biometric systems, rather than only in matching algorithms, by contributing to the further refinement and adoption of the independent test.
Watch the webinar replay here.
In a related development, BixeLab has also had it NVLAP accreditation from NIST renewed for another year, confirming the lab’s conformance with the standards for biometrics testing. BixeLab is still the only lab accredited for biometrics testing to both NIST and FIDO standards in the Southern hemisphere, according to the LinkedIn announcement.