Key takeaways
- Verification: "Is this the same person?" Recognition: "Who is this?"
- 1:N error rates grow with gallery size; 1:1 doesn't have that problem.
- The EU AI Act treats remote identification as high-risk but excludes pure verification.
- Most onboarding needs verification plus liveness, not recognition.
The core difference
| Verification (1:1) | Recognition (1:N) | |
|---|---|---|
| Question | Same person? | Who is this? |
| Inputs | Two images | One image + a gallery |
| Typical use | KYC, re-authentication | Access control, deduplication |
| SIMSIM endpoint | /v1/verify | /v1/recognize |
Error rates you should know
False match rate (FMR) is how often two different people are wrongly matched. False non-match rate (FNMR) is how often the same person is wrongly rejected. They trade off through your threshold. In 1:N search, every extra person in the gallery is another chance for a false match, so identification needs stricter thresholds as galleries grow.
Choosing the right one
- Checking a new customer against their ID? Verification.
- Letting enrolled staff through a door? Recognition against a small, consented gallery.
- Finding duplicate accounts? Recognition across your customer templates.
- Re-authenticating a login? Verification against the enrolled template.
How regulators see them
Under the EU AI Act, remote biometric identification systems are high-risk, while biometric verification that only confirms a claimed identity is excluded from that category. GDPR still applies to both. In the US, BIPA applies to both whenever biometric identifiers are collected. See Western Europe and United States.
