Glossary · Technology
One-to-many matching (1:N)
One-to-many matching, written 1:N, compares one probe face against a database of many enrolled faces and returns the closest candidates. It is the operation behind face search engines, police mugshot searches and watch-list screening. It is distinct from one-to-one verification (1:1), which checks a face against a single claimed identity, and its error rates grow with the size of the database.
Updated November 6, 2026 · 2 min read
How is it different from verification?
| Verification (1:1) | Identification (1:N) | |
|---|---|---|
| Question | Is this the person they claim to be? | Who, among many, does this face resemble? |
| Input | One probe, one claimed identity | One probe, a gallery of N |
| Output | Accept or reject | A ranked candidate list with scores |
| Example | Phone unlock, airport e-gate | Face search engine, mugshot search, watch list |
| Error growth | Fixed per comparison | False matches rise with N |
How does a 1:N search run?
- The probe photo goes through face detection and alignment.
- A face embedding is computed for the probe.
- The embedding is compared with every embedding in the index, using approximate nearest-neighbour search so that millions of comparisons take milliseconds.
- Candidates above a similarity score threshold are returned in rank order; a face search engine shows them with the page each came from.
Why do errors grow with the gallery?
Every enrolled face has a small chance of scoring above the threshold against an unrelated probe. Multiply that by millions of faces and a search returns some strangers who happen to look alike, the doppelganger effect. NIST's identification tests report false-positive identification rates at fixed gallery sizes for this reason, and the same algorithm that is near-perfect at 1:1 can return several wrong candidates at 1:N over a web-scale gallery. The false positive entry covers the measures.
Open-set and closed-set
A closed-set search assumes the person is in the gallery and returns the best candidate. An open-set search, which is what a web face search is, must also allow for the person not being there at all, so the threshold matters more than the rank: the top candidate for an unindexed person is always a stranger. Treating rank one as an identification is the most common misuse of 1:N results.
Where is it regulated?
The EU AI Act bans untargeted scraping of facial images to build recognition databases and restricts real-time remote biometric identification in public spaces; both are 1:N uses. Consumer face search engines are the civilian version of the same operation, and the Clearview and PimEyes cases under GDPR concern the gallery, the index of faces built without consent, rather than the search itself.
Questions
Is a face search engine doing identification?
Technically it runs a 1:N search, but it returns pages rather than identities. Whether a page names the person is a property of the page, not the engine. Engines describe their output as similarity, not identity, for this reason.
Does a bigger index mean better results?
More coverage, and more look-alikes. Precision depends on the threshold and the embedding model as much as on size.
Sources
- 1.NIST FRTE 1:N identification track (checked November 6, 2026)
- 2.ISO/IEC 2382-37:2022, Information technology, Vocabulary, Part 37: Biometrics (checked November 6, 2026)
Entry reviewed November 6, 2026. Found an error? Tell us.
Related entries
- Term Face embedding Face embedding defined: the numeric vector a network computes from a face, why same-person vectors sit close, how it is compared, and why it is biometric data.
- Term Similarity score Similarity score defined: the number from comparing two face embeddings, what it does and does not mean, thresholds, and why a high score is a lead, not proof.
- Term False positive False positive in face recognition defined: a reported match between different people, its relation to false negatives and thresholds, and how rates are measured.
- Term Doppelganger effect The doppelganger effect: why a face search returns look-alikes, how index size and photo quality make it worse, and how to verify a match with non-face evidence.
- Concept Accuracy, bias and testing How face recognition accuracy is measured, what NIST FRTE reports found on error rates and demographic differences, and why marketing claims cannot be checked.
- Law EU Artificial Intelligence Act and the ban on scraping faces EU AI Act Article 5(1)(e): the ban on untargeted scraping of facial images for recognition databases, in force since February 2025, fines up to 7% of turnover.
Cite this entry
FaceSearch Wiki editors. "One-to-many matching (1:N)". FaceSearch Wiki, updated November 6, 2026. https://facesearch.wiki/glossary/one-to-many-matching/