Facial Liveness Bias Testing

ID R&D commissioned the first-ever evaluation of facial liveness detection for demographic bias. The test was performed and documented by BixeLab, a NIST NVLAP-accredited biometrics lab and found IDLive Face to be unbiased. 

Bias is not limited to biometric matching algorithms. It also applies to presentation attack detection (“PAD”), or “liveness detection”.

ID R&D’s fully passive, single-frame facial liveness detection solution, IDLive® Face, was demonstrated to be fair and unbiased. The test resulted in a comprehensive report that documents the methodologies of the test, and also the results. The report is available upon request from ID R&D.

Letter of Confirmation
IDLive Face Exhibits Fairness

The BixeLab Bias Testing Process

The primary goal of the assessment was to determine if the efficacy of liveness detection controls for bona fide users is performing within expectations with similar performance across the measured demographic groups of age, gender, and race. BixeLab concludes that the IDLive Face measured to have a fair liveness detection performance across all demographics in the analysis set containing high-quality images. This is true for a confidence level of 95% using the methodology discussed in this report.

Responsible AI Methodologies

Bias must be actively addressed for both comparators and for liveness. Bias is particularly pronounced for facial biometrics. Responsible AI principles guide the implementation of methodologies that remove bias. Independent third-party validation proves that applying these principles results in a system that is fair and generally free of bias.

ID R&D’s neural networks are trained on extremely large amounts of data. By ensuring adequate representation from all demographics, ID R&D algorithms perform nearly equally across all races and genders.

Read the White Paper

ID R&D authored a white paper to share lessons learned from implementing Responsible AI principles and testing algorithms for bias. The paper provides details about methodologies for design of AI-based algorithms to reduce bias, and about testing of algorithms for bias.

Watch the Webinar

ID R&D participated in a webinar to discuss the liveness detection testing effort and results.

Alexey Khitrov, CEO of ID R&D, was joined by Ted Dunstone, CEO of BixeLab.
The webinar was hosted and mediated by Chris Burt of Biometric Update.

Learn how to stop spoofing and improve security with IDLive Face passive, unbiased facial liveness for presentation attack detection.

If you have questions or would like a demo, contact us.