Technology · Embedded capability

Fingerprint, the modality registries are built on.

Contact platens and standard smartphone cameras, NFIQ 2 quality gating and interoperable templates — so the print captured at enrolment still matches a decade of verifications later.

Talk to the Biometrics Team Contact vs Contactless

NIST PFT III & MINEX III evaluated

Independent 1:1 and interoperable template testing

Contact and contactless

FAP-rated platens and standard smartphone cameras

The short answer

Fingerprint is really two separate questions — how the print is captured, and what the comparison runs against.

Capture is contact or contactless — a finger pressed onto a certified platen, or photographed by a camera. Comparison is 1:1 or 1:N — one print checked against one record, or one print searched across an entire enrolled population.

Axon runs both capture paths and both comparison modes from one integration, with a quality gate in front of each. Policy picks the combination for the journey: a full slap enrolment searched 1:N before a national ID is issued, or a single finger verified 1:1 on a handset with no network at all. Fingerprint latents — the partial prints recovered at a crime scene — are a separate, forensic discipline covered under Criminal ABIS, not this page.

National IDVoter registrationSIM registrationBanking KYCDeduplicationField enrolment

Field operator using an Axon biometric device for fingerprint enrolment in a remote location — Field enrolment — assisted capture in remote conditions

Axon desktop biometric scanner prepared for slap fingerprint capture and document scanning — Dedicated capture station — fingerprint, document and workflow in one device

Finger on a platen, or a camera

Contact is the baseline. Contactless extends the reach.

The two capture modes aren't interchangeable — they trade hardware dependency for measured accuracy in opposite directions, so plan for the modality the destination registry was actually built on.

Finger on a certified platen

Contact capture

The finger is pressed onto an optical or capacitive sensor and imaged directly at 500 ppi. Ridge contact is physical, so scale is fixed by the platen rather than inferred from a camera. Almost every national registry, ABIS and law-enforcement enrolment workflow in production today was built on contact prints — and contact imagery is what the public accuracy benchmarks are principally measured on.

Captures

Direct ridge contact at 500 ppi

Assurance

Highest — the benchmark baseline

Requires

A dedicated FAP-rated sensor

Interchange

WSQ image · ISO/IEC 19794-2 template

Best fit

National ID and civil registry enrolment
Tenprint, rolled and slap capture
1:N deduplication against an existing registry
Anywhere a FAP-rated device is already deployed

Trade-off: contact capture depends on the physical condition of the finger presented. Worn, dry, wet, scarred or henna-marked ridges push up failure-to-capture rates, and a shared platen needs cleaning between subjects.

A camera, no platen

Contactless capture

One or more fingers are photographed — including on the rear camera of a standard smartphone — and the ridge pattern is segmented, scaled and normalised into a comparable print. Nothing is touched, several fingers can be captured in a single frame, and NIST now includes non-contact imagery in its PFT III evaluation, so the accuracy question is measured rather than only demonstrated.

Captures

Camera image, normalised to a print

Assurance

Strong, still maturing

Requires

Camera resolution, focus and light

Interchange

Same template formats as contact

Best fit

Agent and self-service capture with no dedicated hardware
High-volume, low-friction verification
Hygiene-sensitive and public-facing environments
Extending reach where FAP devices are not deployed

Trade-off: contactless prints compare best against other contactless prints. NIST research on cross-matching a contactless capture against a legacy contact-based database shows measurably lower accuracy, from the perspective and elastic distortion a camera introduces — plan for the modality the registry was actually built on, not just the one that is easiest to capture.

Axon contact fingerprint capture device with document bed and biometric scanner — Contact capture station — scanner, document bed and guided capture display

Two captured fingerprint impressions showing ridge detail used for biometric matching — Ridge detail — quality checked before template extraction

A note on mixing modalities. Contactless capture is not a drop-in substitute for a contact print. If the enrolled population was captured on platens, a contactless verification against it will lose accuracy — keep the modality consistent within a programme, or use contactless to extend reach rather than to replace the baseline.

How many fingers, in what order

Single capture

One finger, one impression

The fastest capture, and the usual choice for 1:1 verification at a point of service. Policy normally names the finger — right thumb and right index are common — so the same finger is presented every time and the comparison stays like-for-like.

4-4-2 slap

Ten flat prints in three captures

Four fingers of one hand pressed together, then four of the other, then both thumbs — ten plain impressions in three captures on a FAP60-class platen. Capturing in groups lets the system run a sequence check confirming the fingers were recorded in the right order.

Rolled

Nail-to-nail, one finger at a time

The finger is rolled across the platen to record the full ridge area, including the edges a flat press misses. Slower and more operator-dependent, but it is what a full tenprint submission requires when maximum ridge detail matters — a 4-4-2 slap alone gives flat impressions, not rolled ones.

Which sequence a programme needs is a device-selection decision in its own right — the FAP20, FAP30 and FAP60 device selection guide works through it.

Created once, proven for years

Enrol once. Verify for the life of the identity.

Enrolment and verification aren't the same operation done twice — they trade capture effort for accuracy in opposite directions, and a national programme needs both.

Enrol · Registering the identity

“Who is this person — and are they already in the system?”

The one-time capture that creates the biometric record. Quality is held to the highest bar here, because every verification for years to come is measured against what was captured in this moment. Enrolment typically records more fingers than any single verification will need, runs a 1:N search against the existing population to catch a duplicate registration, and stores both the image and the template.

Multi-finger or 4-4-2 slap capture on a FAP-rated device

NFIQ 2 quality gate before the record is accepted

1:N search against the registry before an identity is issued

Verify · Proving it again

“Is this the same person as the record already on file?”

A single 1:1 comparison against a stored template — seconds long, repeated for the life of the identity. One or two named fingers are usually enough. Because it runs against a template rather than the original image, the same comparison can run on the handset, at a branch, or against a central registry.

One or two named fingers at the point of service

1:1 against a card, a device or a central record

A pass, retry or refer decision inside the journey

Enrol once, verify forever

A national ID enrolment happens once and takes minutes. The verifications it enables — a SIM purchase, a grant collection, a bank login — happen millions of times over the decade that follows. That asymmetry is the whole argument for spending the extra capture time at enrolment.

National ID enrolment

Full slap capture, quality-gated at the desk and searched against the population register before a credential is issued.

Enrol · 1:N

Election list deduplication

Clean the voter roll before polling, so one person cannot appear twice under two identities.

Enrol · 1:N

SIM registration

Bind the subscriber to a biometric record at activation, and search the existing base for a prior registration.

Enrol · 1:N

Polling-day voter verification

Confirm the voter against their enrolled record at the station, on the device, with no network dependency.

Verify · 1:1

Banking and mobile money onboarding

Match the applicant against the national ID record so the account opens against a verified identity, not a claimed one.

Verify · 1:1

Benefit and grant disbursement

Prove the recipient is present at collection — the control that removes ghost beneficiaries from a payroll.

Verify · 1:1

What gets checked, and against what

Quality gating and the standards behind interoperability.

A print is only as useful as the checks run on it before it's stored, and only as portable as the format it's stored in — three different layers, often confused as one.

Axon tablet showing a live fingerprint verification screen with a matching quality score — Quality score shown live — matching confidence checked before a result is returned

Axon rugged tablet with integrated fingerprint reader for compact enrolment and verification — Compact field device — FAP fingerprint capture for assisted workflows

Check	Contact	Contactless	What it means
NFIQ 2 quality scoring	Calibrated	Indicative only	NFIQ 2 scores a print 0–100 for how much it is likely to contribute to a successful match. It is calibrated on 500 ppi contact images, and NIST research found its scores are a weaker predictor of match accuracy on some contactless captures — so contactless needs its own thresholds, not borrowed ones.
Smudge and moisture detection	Yes	Not applicable	A wet, greasy or smeared platen press blurs ridge boundaries and produces a record that will fail later. Contactless capture has no platen to smudge, but trades that problem for focus and motion blur instead.
Fingertip and partial capture	Yes	Yes	Confirms the core of the print sits inside the capture area and that enough ridge area was recorded to support a reliable comparison — rejected at the desk, not discovered years later.
Sequence check on slap capture	Yes	Yes	Segments a multi-finger slap into individual fingers and validates they were captured in the expected order — the control that stops a left hand being filed as a right.
Presentation attack detection	Sensor & software	Software only	A contact sensor can add hardware signals alongside image analysis. A camera-based capture has only the image to work from, which makes contactless PAD the less mature of the two today.
Matching against a legacy registry	Yes	Reduced accuracy	Almost every registry in production was built from contact prints. Cross-matching a contactless capture against it is measurably less accurate than a like-for-like comparison.

Fingerprint images

What the sensor produced.

WSQ: The FBI’s wavelet compression for 8-bit, 500 ppi prints, developed with NIST and Los Alamos. It exists because ordinary JPEG at the ratios needed for national-scale storage introduces blocking artefacts that destroy the fine ridge detail a matcher depends on.
JPEG 2000: What the FBI specifies in place of WSQ for 1000 ppi imagery, where WSQ does not apply.
Raw and PNG: Uncompressed or lossless. Used for local capture feeding immediate template extraction, and for archival masters where nothing may be discarded.
ISO/IEC 19794-4: The international finger image interchange record.

Minutiae templates

What the matcher actually compares.

ISO/IEC 19794-2: The international finger minutiae format — a compact, non-image record of ridge endings and bifurcations, rather than a picture of a finger.
ANSI/INCITS 378: The US national minutiae format, and the one NIST MINEX III tests for cross-vendor interoperability.
Proprietary template: A vendor-defined encoding, more accurate than an interoperable one and readable only by its own matcher — a deliberate trade, not a default.

Transactions and quality

How it travels between systems.

ANSI/NIST-ITL 1-2011: The envelope standard for exchanging biometric records between agencies and into an ABIS — Type-14 carries the fingerprint image, Type-9 the minutiae data.
ISO/IEC 29794-4: The finger image quality standard NFIQ 2 implements, so a quality score means the same thing to whoever receives it.
ISO/IEC 39794-4: The next-generation finger image format. ICAO inspection systems have had to read it since January 2026, with issuance moving to it by 2030 — noted here as industry context, not a stated Axon capability.

Independently evaluated, not self-declared

NIST PFT III

Proprietary-template 1:1 verification

NIST’s ongoing evaluation of proprietary fingerprint template generation and one-to-one matching, across rolled, plain, ink, 500 and 1000 ppi and non-contact imagery. It replaced PFT II, which was retired in 2019.

NIST MINEX III

Interoperable ANSI/INCITS 378 templates

Tests whether one vendor’s minutiae template can be read and matched by another vendor’s matcher — the compliance route for the US Government PIV programme, and the reason an interoperable template can be trusted at all.

NFIQ 2

Image quality, calibrated to match utility

An open NIST tool that scores a fingerprint 0–100 for its expected contribution to matching accuracy — not a matcher benchmark, but a way to reject a poor capture before it becomes a permanent record.

NIST publishes ordered results for both evaluations, from every submitted vendor — it does not certify or endorse a product. See the FAQ for our current standing.

Benchmarks measure algorithms, not programmes

PFT III and MINEX III test matching software against fixed datasets. Field accuracy is set by the sensor, the operator, the lighting, and the condition of the fingers presented.

Some fingers will not enrol

Worn, scarred, dry, henna-marked and amputated fingers happen more often than vendor material admits, and a percentage of any population will not produce a usable print. A credible programme plans an exception path — an alternate finger, a supervised override, or a second modality — before deployment.

A score still needs a policy

Match, no-match and uncertain each need a defined next step: pass, re-capture, try another finger, step up to a second modality, or refer to an operator.

On the device, or against the registry

Where the comparison runs is a deployment decision.

Local matching answers 'is this the person on this card, right now, with no network.' Central matching answers a different, harder question: 'has this person registered before, under any name.'

On device

Local 1:1 matching

“Can this decision be made here, right now, offline?”

The template is held on the device, on a smart card or in a secure element, and the comparison runs where the finger is presented. Nothing biometric leaves the handset, the result returns in milliseconds, and the workflow keeps running when the network does not. In its strongest form — match-on-card — the template never leaves the chip at all.

Works fully offline at the point of service

Biometric data stays on the device or the card

Bounded to the records actually held locally

Server side

Central 1:1 and 1:N matching

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

The capture is sent to a central matcher — a 1:1 check against the registry’s record for a claimed identity, or a 1:N search across the whole enrolled population to find a record nobody claimed. Only the server side can answer the second question; at national scale that is an ABIS function, with binning, ranked candidate lists and adjudication behind it.

The only place a 1:N search can run

One authoritative record, centrally governed

Every comparison logged and auditable

Axon handheld device showing a biometric voter authorisation screen with a fingerprint sensor — Verification at the point of service — fingerprint sensor, live decision, no network dependency

Fingerprint liveness

A matcher answers whether two prints came from the same finger — it has nothing to say about whether the thing on the sensor was a finger at all. Presentation attack detection is the separate control that does, tested and reported under ISO/IEC 30107-3 as APCER and BPCER, against gelatin and silicone casts, printed ridge patterns and worn overlay films — and, for contactless capture, a high-resolution photograph or a screen.

How Axon approaches liveness

Capture

Contact platen or camera; single finger, 4-4-2 or rolled

Assess

NFIQ 2 score, smudge, fingertip and partial checks

Template

Proprietary or ISO/IEC 19794-2 minutiae record

Match & decide

1:1 on device or server, 1:N in the ABIS

National ID

Slap enrolment, quality-gated and deduplicated against the population register before a credential is issued.

Election lists

Deduplicate the voter roll before polling, then verify voters at the station — offline where the network is not.

SIM registration

Bind the subscriber to a biometric record at activation, and re-verify them at swap, port and recovery.

Banking verification

Match an applicant against the national ID record so the account opens against a verified identity, not a claimed one.

Start with the enrolment

Let us design the capture policy before the matcher.

We'll work from the population, the field conditions, the devices already deployed and the registry you have to match against — then set the finger set, the quality gate and the matching topology around them.

Talk to the Biometrics Team

Common questions

Fingerprint biometrics, without the fog.

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

What is the difference between fingerprint enrolment and verification?+

Enrolment is the one-time capture that creates the biometric record — usually multiple fingers, at the highest quality the device can produce, searched 1:N against the existing population to catch a duplicate registration. Verification is the repeated 1:1 comparison against that stored record, usually one or two fingers, seconds long. Enrolment happens once; the verifications it enables happen for the life of the identity, which is why enrolment quality is worth the extra capture time.

What does 4-4-2 fingerprint capture mean?+

A slap capture sequence: the four fingers of one hand pressed flat together, then the four fingers of the other hand, then both thumbs — ten plain impressions in three captures, on a FAP60-class platen. Capturing in groups lets the system segment the slap into individual fingers and run a sequence check confirming they were recorded in the right order. Note that 4-4-2 produces flat impressions, not rolled ones — a full tenprint submission normally requires rolled captures as well.

Is contactless fingerprint capture as accurate as a contact scanner?+

Not yet, and the honest answer depends what you are matching against. Contactless prints compare well against other contactless prints. Cross-matching a contactless capture against a legacy contact-based registry is measurably less accurate, because a camera introduces perspective and elastic distortion a platen does not. NIST includes non-contact imagery in the PFT III evaluation and publishes guidance on evaluating contactless devices, so this is a measured gap rather than a matter of opinion — and capturing several fingers at once recovers some of it.

What is NFIQ 2 and what does the score mean?+

NFIQ 2 is an open NIST tool that scores a fingerprint image from 0 to 100 for its utility — how much that sample is expected to contribute to a successful match. 0 means no utility, 100 the highest. It is not the same scale as the original NFIQ, which used 1 to 5 with 1 as the best, so the direction is inverted between the two. Its features are standardised in ISO/IEC 29794-4, and it is calibrated on 500 ppi optical and ink contact prints — which is why contactless captures need their own thresholds.

What is WSQ, and when would you use it instead of a raw image?+

WSQ — Wavelet Scalar Quantization — is a lossy compression format the FBI developed with NIST and Los Alamos specifically for 8-bit, 500 ppi fingerprint images. It exists because standard JPEG at the ratios a national registry needs introduces blocking artefacts that destroy the fine ridge detail matching depends on. Use WSQ for storage and exchange at scale — into an ABIS, or between agencies. Use raw or a lossless format for local capture feeding immediate template extraction, and for archival masters. For 1000 ppi imagery the FBI specifies JPEG 2000 instead.

What do NIST PFT III and MINEX III actually measure?+

Two different things. PFT III evaluates proprietary fingerprint templates in one-to-one verification — vendor-defined encodings only their own matcher can read, tested across rolled, plain, ink, 500 and 1000 ppi and non-contact imagery. MINEX III evaluates interoperable templates in the ANSI/INCITS 378 format, testing whether one vendor’s template can be matched by another vendor’s matcher. Proprietary templates are more accurate; interoperable templates let you change vendor — a strong result in one implies nothing about the other. NIST publishes ordered results for both, and it does rank submissions on its published metrics; what it does not do is certify or endorse a product.

Can a fingerprint be spoofed, and what stops it?+

Yes. Casts made from gelatin, wood glue, silicone or latex, thin films worn over a genuine finger, and printed ridge patterns are all documented attacks — and a camera-based contactless capture adds photographs and screens to that list. The countermeasure is presentation attack detection, tested and reported under ISO/IEC 30107-3 using two figures: APCER, the rate attacks get through, and BPCER, the rate genuine users are wrongly rejected. Either number can look good alone, so ask for both. Contact sensors can combine hardware signals with image analysis; contactless capture has only the image, which makes its PAD the less mature of the two today.

What happens when someone’s fingerprints will not enrol?+

It happens more often than vendor material admits. Manual labour, age, dryness, scarring, henna and certain medical conditions all degrade ridge detail, and a percentage of any national population will not produce a usable print. A credible programme plans for this before deployment: try an alternate finger, allow a supervised exception with a recorded reason, and fall back to a second modality such as face where the policy calls for it. Designing the exception path is a policy decision, not an algorithm problem.

Where does Axon’s fingerprint algorithm rank in NIST testing?+

CONFIRM RANK in CONFIRM TRACK, as of CONFIRM DATE — see the published NIST PFT III results directly. NIST publishes ordered results for every submission; it does not certify or endorse a product, and standings move with each new submission round, so treat any dated ranking as a snapshot, not a permanent status.