Ant Group’s Biometric Security Testing Lab: Automated Detection and Evaluation of Fingerprint and Face Recognition Systems

Biometric identification (face, fingerprint, voice) is now ubiquitous in both offline transactions and online services, yet traditional password‑based methods are cumbersome and vulnerable to forgery and cracking. Ant Group’s Ant Security Tianji Lab has built a comprehensive biometric security testing system to evaluate and improve the safety of these technologies.

Background – Industry‑wide biometric security testing suffers from high labor costs, long cycles, lack of quantification, and fragmented standards, limiting large‑scale adoption. Ant Security Tianji Lab has contributed to national standards (e.g., IT Mobile Device Biometric Recognition Part 8), IIFAA requirements, and IEEE P2884/P2891, establishing unified evaluation criteria.

Evolution of Detection

1.0 Manual Detection Era : Relied on expert experience; unable to scale or provide statistically meaningful, standardized results.

2.0 Automated Detection Era : Introduced a fully automatic biometric security evaluation system (2020) with >200,000 test cases per device, millimetre‑level precision, and zero human intervention, enabling quantifiable and reproducible outcomes.

3.0 Intelligent Adversarial Detection Era : Combines AI‑driven attack generation, computer vision, and robotics to assess anti‑spoofing, model robustness, link security, and privacy compliance, defending against deep‑fake, presentation, and injection attacks.

Fingerprint Recognition Security

The lab outlines the fingerprint acquisition pipeline (capacitive, optical, ultrasonic), preprocessing, feature extraction, and matching. In 2019 it released the IIFAA Biometric Fingerprint Security Test Requirement, defining multi‑dimensional metrics that have become the de‑facto standard for >80% of Android devices.

Performance metrics focus on FAR (False Acceptance Rate) and FRR (False Rejection Rate), targeting FRR ≤ 3 % and FAR ≤ 1/50 000 across varied scenarios (cold environments, post‑hand‑wash, etc.).

Presentation‑Attack Materials

Two categories of spoof materials are used: 2D (printed fingerprints) and 2.5D (soft‑material embossed fingerprints). Over 300 formulations were tested, selecting ~20 effective samples. The Security Attack Ratio (SAR) is calculated for each material to quantify resistance.

Non‑Fingerprint Texture Detection

Seven common non‑fingerprint textures (e.g., screen protectors, bubbles) are defined to ensure sensors can distinguish real fingerprints from foreign patterns.

Intelligent Adversarial Fingerprint Testing

Robotic arms equipped with vision and reinforcement‑learning‑based path planning automatically locate, press, and record results for thousands of spoof samples, achieving sub‑2 mm precision and fully reproducible test reports.

Face Recognition Anti‑Spoofing

Presentation attacks include 2D media (photos, videos) and 3D media (masks, head‑models). The lab built an automated system that selects attack materials, identifies vulnerable points, executes attacks with robotic arms, and generates real‑time SAR statistics.

Environmental lighting (50–100 000 lux, 2700–5500 K) is automatically calibrated using a CL200A lux meter, with closed‑loop control to mimic diverse real‑world conditions.

Full‑Chain Trusted Evaluation

The lab offers a holistic assessment covering model robustness, link security, privacy compliance, and anti‑spoofing completeness, aligned with both domestic and international standards. Over 70 % of Android devices worldwide are evaluated using this framework.

Robotic Transport and 24/7 Operation

A composite testing robot integrates a mobile base, liftable robotic arm, radar, and multi‑camera SLAM for indoor localization and map reconstruction, enabling continuous, unattended testing across multiple test stations.

Hardware diagrams, system architecture, and performance charts are illustrated throughout the original article (images retained below).

Overall, Ant Security Tianji Lab provides a scientifically grounded, AI‑enhanced, fully automated biometric security testing platform that supports large‑scale, reproducible, and 24/7 unattended evaluation of fingerprint and face recognition systems.