Face Recognition Trials

13/1/2026

POC  Aims

1. Benchmark biometric systems from multiple vendors under
controlled, repeatable conditions.
Focuses on critical aspects such as 1:1 verification (e.g., kiosk‑to‑
passport matching) and 1:N identification (e.g., bag‑drop‑to‑gallery matching)

2. Assess
the impact of image quality on matching accuracy using the Open Face Image Quality (OFIQ)
standard.

3. Assess PAD algorithms from the vendors (passive)

Summary

• Each algorithms generated approximately 142k records, corresponding to 7.1k transactions compared against 7.7k enrolled templates to form the candidate
  galleries. 

Matching Accuracy

Kiosk to Bag Drop (1:N)

Bag Drop to Passport (1:N)

Kiosk to Bag Drop Speed

Rank 1 Identification Rates (against Passport)

Vendor Comparison

Vendor Comparison - PAD

Quality of Passport Image

Quality of Kisok vs Bagdrop

Bona fide transaction tests (10 tests) completed successfully in all cases, with no failures recorded. Presentation attack results varied by PAI species: 

• Printed photo attacks on paper were unsuccessful in 9 of 10 cases. 

• Digital display attacks using mobile devices were unsuccessful in all ten cases. 

• Balaclava-style wearable artefacts resulted in successful bypass in all ten test cases. 

• T-shirt presentation attacks resulted in 9 successful bypasses and 9 rejections (18 tests). 

• Silicone prosthetic attacks were successful in 5 cases and unsuccessful in 4 (9 tests). 

• Silicone mask tests, conducted end-to-end, resulted in 4 successful and 1 unsuccessful outcome (5 tests). 

PAD Trials at Amadus

PAD Trials at Amadus

Next Steps

• Vendor Feedback
• Biometrics and Identity Strategy
• Ongoing Risk and Vulnerability Analysis
• Publishing Outcomes