Face search1:N

Checking a new face against every face already in your system, either your own user base (to catch someone opening a second account under a different name), or an external watchlist of known fraudsters (to catch a repeat offender).

Instead of asking whether two faces match, it asks whether this face matches any face on file.

Comparing one face against every face in an enrolled gallery, whether that's your own user base or an external watchlist, returning the closest matches ranked by similarity score rather than a single yes-or-no answer like a one-to-one match. A gallery is the collection of enrolled faces the search runs against.

A face search typically works like this:

1. The system converts the new face into a template.
2. It compares the template with every template in the gallery.
3. It returns the closest candidates ranked by similarity score.
4. A threshold or a human reviewer decides whether any candidate is a true match.

Services commonly search two kinds of gallery:

• Their own user base, to find one person who holds several accounts under different names.
• An external watchlist, to catch a known fraudster who tries again under a new name.

Face search sits next to terms that are easy to mix up:

• Face match (1:1) compares one face with one other, instead of one with many.
• Duplicate detection finds that a new applicant is already in the system, which face search supports.
• Fraud ring is a coordinated group of linked identities, which repeated faces across accounts can reveal.