Unidata Expanded Biometric Training Dataset Representation
The expanded dataset reduces demographic accuracy gaps in liveness detection systems by training against diverse mask attacks.
Updated on Oct. 14, 2026 in Cybersecurity
Live Poll
Does more diverse training data make you trust biometric liveness systems more?
Unidata has expanded its commercial biometric liveness detection training dataset to include 53,870 videos featuring 50 participants from African, Latin American, and Asian backgrounds. This research effort aims to correct uneven error rates found in existing biometric systems.
Why it matters
Systematic testing previously revealed that liveness detection error rates are unevenly distributed across demographic groups. Expanding training data is a necessary step to reduce these accuracy disparities in commercial security deployments.
Researchers at Carnegie Mellon University Africa leveraged a dataset of 53,870 videos to reduce demographic accuracy gaps from 3.07 percentage points to 0.75. Previous industry benchmarks had shown a maximum error rate of 5 percent on customer systems.
The players
Unidata
A developer of biometric liveness detection technology focused on commercial identity verification systems.
Carnegie Mellon University Africa
A research institution based in Rwanda that focuses on engineering and technology solutions tailored for regional and global challenges.
The details
The research methodology required 50 participants to perform head movements and presentation attacks, including the use of custom silicone masks created from individual plaster casts. To prevent rejection based on color boundaries, visible skin areas were specifically matched to mask tones. These techniques allow algorithms to better distinguish between human subjects and sophisticated spoofs like printed photos or phone replays.
Timeline
2023: NIST evaluated passive biometric liveness detection algorithms.
The Tech Race
The 2023 NIST evaluation of passive biometric liveness detection algorithms highlighted significant imbalances in subject representation, featuring 15,131 images of White subjects compared to 3,583 of Black subjects and 1,400 of Asian subjects. This current effort seeks to bridge that gap by diversifying the underlying training data used for commercial identity verification.
Users of biometric security systems may experience fewer false rejections as these training advancements are integrated into commercial software. The update primarily affects the reliability of identity verification workflows that utilize liveness detection to prevent spoofing.
The takeaway
Closing the demographic accuracy gap in liveness detection is essential for creating equitable and reliable authentication systems. Watch for future NIST evaluation updates to see if these expanded datasets lead to standardized improvements across the broader biometric industry.
Further reading
For more on how identification technologies evolve to meet security standards, explore our Cybersecurity section.
Source note: This article includes information reported by Biometric Update.
Live Poll
Does more diverse training data make you trust biometric liveness systems more?
