Technology · Embedded capability

Face matching, proof it’s the same person.

Axon combines NIST-benchmarked 2D matching with 3D face matching, and both 1:1 verification and 1:N identification — so the right comparison runs against the right evidence, every time.

Design a Matching Policy Compare 2D and 3D

NIST FRVT benchmarked

Independently tested 1:1 & 1:N accuracy

3D-to-3D matching

Against a live FaceMap, not a flat photo

The short answer

Face matching is really two separate questions — which evidence, and against how many candidates.

Evidence is 2D or 3D — a flat image compared against another flat image, or a live 3D FaceMap compared against a prior 3D enrolment. Mode is 1:1 or 1:N — one face checked against one record, or one face searched against millions.

Axon runs both engines and both modes from a single integration, and policy selects the right combination for the journey — a fast 1:1 2D check at onboarding, a 1:N search to catch a duplicate registration, or a 1:1 3D re-authentication when the risk of the transaction justifies it.

Identity verificationSIM registrationDeduplicationWatchlist screeningAccount recovery

Two kinds of evidence

3D and 2D matching both have a place.

2D is not a fallback option; it is the only option whenever a document photo is involved. 3D adds the strongest available match evidence once a 3D enrolment exists.

3D FaceMap vs 3D FaceMap

3D face matching

Matches a live 3D FaceMap — the depth and texture data captured during a 3D liveness session — against another 3D FaceMap from a prior enrolment. Because both sides of the comparison are live biometric captures rather than a flat image, it is far harder to defeat with a printed photo, screen replay or 2D mask.

Compares

3D FaceMap vs 3D FaceMap

Assurance

Highest, spoof-resistant

Requires

A prior 3D enrolment on file

Processing

Online, real-time capture

Best fit

Re-authentication against a prior 3D enrolment
Account recovery and SIM swap
High-value repeat transactions
Journeys where a live capture is already justified

Trade-off: 3D-3D matching only works once someone has a 3D enrolment on record. It cannot be run against a passport or ID card photo — those are inherently 2D.

Any camera, any photo

2D face matching

Converts each face — a live selfie and a passport, ID card or previously enrolled photo — into a 2D template (a facial embedding) and compares the two templates. This is the industry-standard approach benchmarked by NIST FRVT, and the only option available whenever one side of the comparison is a printed or scanned document photo.

Compares

2D image vs 2D template

Assurance

Strong, NIST-benchmarked

Requires

Any facial image, live or archived

Processing

Edge, offline or server

Best fit

Selfie-to-document verification at onboarding
1:N deduplication against an existing photo archive
Watchlist and sanctions screening
Offline or lower-connectivity capture

Trade-off: 2D matching alone does not confirm the input image was captured live. Pair it with liveness detection wherever presentation-attack risk matters.

A note on document verification. Any check against a passport, ID card or driving licence photo is a 2D comparison, because the document photo itself is a flat image. 3D-3D matching applies to re-authentication against a prior 3D enrolment, not first-time document checks.

Two matching modes

Verify one. Search millions.

1:1 and 1:N answer different questions, and a mature identity programme needs both — often in the same registration.

1:1 · Verification

“Is this the same person as this document or enrolment?”

A single, real-time comparison between two faces — a live selfie against a document photo, or against a previous enrolment. Fast enough to sit inline in an onboarding or authentication flow.

Selfie vs passport or ID photo at onboarding

Re-authentication at SIM swap or recovery

Confirming the agent-assisted enrolment matches the applicant

1:N · Identification

“Has this person already registered, under any name?”

One face searched against a database of millions of enrolled identities. Used to find duplicates and matches the person themselves did not disclose — the control that stops a repeat registration under a new identity.

SIM registration deduplication

Ghost-subscriber and synthetic-identity detection

Watchlist and sanctions screening

Both, in one registration

A single SIM registration typically runs a 1:1 check (does the selfie match the document?) and a 1:N check (has this person already registered?) before the subscriber is accepted.

New SIM or account registration

Search the existing subscriber base for a prior registration under any identity before accepting a new one.

1:N · 2D

Onboarding selfie vs ID document

Compare the live selfie against the document photo captured moments earlier in the same session.

1:1 · 2D

SIM swap or account recovery

Re-authenticate against the subscriber’s original 3D enrolment for the strongest available match evidence.

1:1 · 3D

Watchlist and sanctions screening

Search a photo against watchlist and sanctions image sets at scale.

1:N · 2D

High-value repeat transaction

Confirm the same physical person initiated a previous 3D-enrolled session, not just a matching photo.

1:1 · 3D

Legacy photo database cleanup

Deduplicate an archived enrolment database without recapturing every subject.

1:N · 2D

Accuracy & fairness

A score is only as good as what stands behind it.

Independent benchmarking, quality gating and demographic monitoring matter as much as the raw match algorithm.

Consideration

What it means

Independent benchmarking

2DNIST FRVT tested

3DFaceTec evaluated

Both engines are assessed against public, third-party test sets rather than vendor claims alone.

Works against a document photo

2DYes

3DNo — needs prior 3D enrolment

Passport and ID photos are flat images, so document verification is always at least partly 2D.

Resistance to photo or screen spoofing

2DDepends on liveness pairing

3DBuilt into the match itself

3D-3D matching is inherently harder to spoof because both sides of the comparison are live captures.

Demographic equity

2DIndependent of demographics

3DIndependent of demographics

Matching accuracy is validated to perform consistently across demographic groups, rather than varying by race, gender or age.

Quality gating

2DRequired

3DRequired

Poor-quality captures are rejected before matching rather than silently scored.

Benchmarks are a floor, not a ceiling

NIST FRVT is a credible baseline, but production accuracy also depends on capture quality, lighting and the enrolled population.

1:N is a probability game

At database scale even a small false-match rate produces real collisions, so results need adjudication thresholds, not just one score.

A score still needs a policy

Match, no-match and uncertain results all need a defined next step: pass, refer, re-capture or step up.

Built in, not bolted on

One decision inside the identity flow.

Axon integrates capture, matching, liveness and business rules so the customer sees one coherent journey and the organisation receives one auditable outcome.

Capture

Selfie, document photo or 3D FaceMap

Compare

1:1 verification or 1:N search

Score

2D or 3D match, quality and confidence

Decide

Match, no-match, refer or step up

Identity Verification

Face matching combined with liveness, document trust and NFC into one onboarding decision.

Polaris

1:N dedup stops re-registration under a new identity; 1:1 verification confirms the subscriber at swap and recovery.

RealDoc

Document photo extraction feeds the 2D match reference automatically during onboarding.

Start with the journey

Let us map the right matching policy to your channels.

We will work from fraud risk, connectivity, throughput and existing enrolment data — then embed the right combination of evidence and mode into the solution.

Talk to the Identity Team

Common questions

Face matching, without the fog.

Clear answers for product, fraud, security and operations teams.

What is the difference between 1:1 and 1:N face matching?+

1:1 verification answers "is this the same person as this specific document or enrolment?" with a single comparison. 1:N identification searches a face against many enrolled identities to answer "has this person already registered, under any name?" Axon uses both, depending on the journey.

What is the difference between 2D and 3D face matching?+

Both compare against a stored template rather than a raw image, but the template differs. 2D face matching builds a 2D template from a flat facial image — a selfie, passport, ID card or archived photo. 3D face matching builds a 3D FaceMap, captured during a 3D liveness session, and compares it against another 3D FaceMap on file. Because both sides of a 3D comparison are 3D biometric captures rather than a flat image, it is harder to defeat with a printed photo or screen replay.

Can you 3D-match a face against a passport photo?+

No. A passport or ID photo is a flat 2D image, so any comparison against it is a 2D match. 3D-to-3D matching is only possible once a person already has a 3D FaceMap enrolment on file, which is why it is normally used for re-authentication rather than first-time document verification.

Is face matching the same as liveness detection?+

No, but they work together. Liveness estimates whether the presented face belongs to a real person in the moment. Face matching estimates whether that face is the same person as a document or a prior enrolment. Axon runs both checks inside the same identity decision.

How does 1:N matching support SIM registration?+

Every new registration is searched against the existing subscriber base before it is accepted, so the same person cannot register a second identity under a different name — the same control used to catch what regulators call a "ghost subscriber."

Is NIST FRVT certification enough on its own?+

It is a credible independent baseline for algorithm accuracy, not a guarantee of production performance. Real-world accuracy also depends on capture quality, lighting, camera hardware and the demographic makeup of the enrolled population.

Does Axon sell face matching as a standalone product?+

No. Face matching is delivered through IDToolbox, part of the Identity Verification solution, and embedded into platforms such as Polaris — so a match result works alongside liveness, document trust and business rules instead of standing alone.