How does face recognition work across different faces and conditions?
Face recognition performance varies with camera quality, lighting, pose, image age, thresholds and population. FaceTagr's engine is benchmarked by NIST FRTE, but a benchmark is not a guarantee for every deployment. Test representative users and working conditions in a pilot, measure false matches and missed matches, and review exceptions before rollout.
Last reviewed: 18 October 2026 · FaceTagr team
What affects the recognition result?
| Factor | What to test |
|---|---|
| Lighting | Daylight, dim rooms, backlighting and your actual camera position |
| Capture quality | Resolution, motion blur, pose and lens cleanliness |
| Reference images | Image age, enrolment quality and any permitted updates |
| Population | Representative participants and differences in error rates |
| Thresholds | False-match and missed-match rates at your chosen settings |
| Liveness | Genuine users and photo or screen replay attempts |
How should you reduce deployment risk?
Step 1
Review the specific NIST test, task and thresholds rather than a universal accuracy percentage.
Step 2
Run a pilot with representative participants, appropriate permissions and a human fallback.
Step 3
Measure genuine-user completion, false matches, missed matches and exception resolution.
Step 4
Test offline behaviour and later sync on the phones, tablets or CCTV you intend to use.
Plan a pilot around your workflow
For privacy or deletion requests, contact support@facetagr.com. Read our privacy policy.
Sources
Product capabilities describe FaceTagr's supported workflows, not third-party endorsement. This page is general information, not legal advice. Check your obligations with your data protection officer or adviser.
Frequently Asked Questions
Is one accuracy number valid for every population?
No. Results depend on the task, threshold, gallery and image conditions. Do not infer demographic performance from a headline score.
Can it work without a network?
Yes. FaceTagr's mobile SDK supports fully offline 1:1 verification, 1:N identification and passive liveness on Android and iOS. Reporting and integration need a connection when records are synchronised.