1:1 vs 1:N face recognition: which do you need for KYC?

1:1 face recognition (verification) compares one face against one claimed identity to answer "are you who you say you are?". 1:N face recognition (identification) searches one face against a database of many to answer "who is this?". For KYC and identity verification, you almost always want 1:1.

1:1 face recognition (verification)

A 1:1 check takes a live selfie and compares it to a single reference image — typically the portrait read from an identity document or a mobile driver licence. It returns a similarity score, which you threshold for pass, fail, or step-up to review. There is no database of other people involved; the comparison is one-to-one, tied to the identity the person is claiming.

This is the model behind remote onboarding: read the document with ID document OCR, confirm the person is live with liveness detection, then use 1:1 face matching to bind the person to the credential.

1:N face recognition (identification)

A 1:N system searches a probe face against many enrolled faces to find a match — for example, de-duplicating a population or spotting a returning user without a claimed identity. It is powerful but heavier: it needs a curated gallery, raises stronger privacy and governance obligations, and its error profile scales with the size of the database.

Which do you need?

 1:1 verification1:N identification Question answeredAre you who you claim to be?Who is this person? Reference dataOne image (the claimed identity)A gallery of many faces Typical useKYC, onboarding, account recovery, step-upDe-duplication, watchlists, surveillance Privacy footprintLower — no database searchHigher — governance-heavy

For onboarding, age checks, account recovery and most compliance flows, the answer is 1:1. You have a claimed identity and a document; you only need to confirm the person matches it.

Accuracy: FAR, FRR and independent testing

Face matching accuracy is expressed as a trade-off between two error rates:

• False Accept Rate (FAR) — how often the system wrongly accepts a non-match.
• False Reject Rate (FRR) — how often it wrongly rejects a genuine match.

Tightening the threshold lowers FAR but raises FRR, and vice versa — so a single "accuracy %" without a stated threshold and test set is close to meaningless. The most credible reference point is independent benchmarking such as NIST's Face Recognition Technology Evaluation (FRTE, formerly FRVT). Treat any vendor's headline number with caution unless it names the threshold, dataset and an independent evaluator.

Privacy and governance (Australia)

Face data is sensitive personal information. In Australia it sits under the Privacy Act and the OAIC's guidance, so minimise what you collect, be transparent about it, and set a clear retention policy. 1:1 verification is generally easier to justify than 1:N identification because it does not build or search a biometric database — a meaningful point in a privacy impact assessment.

How it pairs with liveness

Face matching alone can be fooled by a photo of the right person. That is why 1:1 matching is paired with liveness detection: liveness proves a live person is present, matching proves it is the right person, and together they support remote KYC. For the presentation- and injection-attack side, see passive vs active liveness detection.

Frequently asked questions

Is 1:1 or 1:N better for KYC?

1:1. KYC has a claimed identity and a document, so you only need to confirm the person matches that one reference — not search a database of everyone else.

What is the difference between face verification and face identification?

Face verification is 1:1 (one-to-one, "is this the claimed person?"); face identification is 1:N (one-to-many, "who is this?"). The terms map directly to the two modes above.

How accurate is 1:1 face recognition?

It depends entirely on the chosen FAR/FRR threshold and the test conditions. Ask for independent benchmarking (e.g. NIST FRTE, formerly FRVT) rather than a bare percentage, and tune the threshold to your risk appetite per flow.

Is face recognition legal in Australia?

Yes, when done in line with the Privacy Act — collect only what you need, be transparent, get consent where required, and set a retention policy. 1:1 verification carries a lighter privacy footprint than 1:N identification. This is general information, not legal advice.