Key takeaways
- It turns a face into a numeric template and compares templates, not photos.
- Verification (1:1) and identification (1:N) are different jobs with different risks.
- A threshold you choose decides what counts as a match.
- Biometric data is regulated almost everywhere — design for consent from day one.
How it works
Every face recognition API runs the same pipeline. First, detection finds the face and its landmarks. Next, alignment rotates and scales it to a standard pose. A neural network then produces an embedding — a list of numbers that captures what makes that face distinctive. Finally, matching compares embeddings: similar faces have nearby vectors.
Because matching happens on templates, a well-designed service doesn't need to keep the original photo. SIMSIM stores templates only.
Verification vs identification
Verification (1:1) asks whether two images show the same person — a selfie against an ID card, say. Identification (1:N) searches one face against a gallery to find who it is. Verification is the backbone of KYC; identification powers access control and watchlists. Regulators often treat identification as higher risk. Our comparison guide goes deeper.
Scores and thresholds
An API returns a similarity or confidence score. You pick a threshold: above it, a match; below it, not. Raise it and you get fewer false matches but more genuine users rejected; lower it and the reverse. Tune it on data that looks like your real traffic.
What people use it for
- Remote onboarding and KYC in fintech and crypto.
- Age assurance and one-account-per-player in iGaming.
- Step-up authentication in payments.
- Staff access control in retail and logistics.
Using it responsibly
Biometric data is sensitive under GDPR, BIPA, PIPEDA, Australia's Privacy Act and India's DPDP Act. Give people clear notice, get consent where required, offer an alternative, keep data only as long as needed and never use it for covert surveillance. Start with our laws-by-region guide.
