The document portrait
Extracted from the government ID during document reading: cropped from the printed photo on a driver's license, provincial card or passport photo page.
Face Match
The face on the card is the face on the phone.
ShieldID compares the portrait printed on a government ID with a live capture taken during the verification, using a face-recognition model with detection and alignment built for real phone photos, not studio portraits. The result is pass or fail against a fixed decision line. We never print a percentage we can't stand behind.
What is compared
Two photos. One question.
Face Match answers exactly one thing: is the person on the phone the same person on the document? It is a one-to-one comparison, never a search against a database of other faces.
The live capture
Taken on the applicant's own phone during the guided liveness challenge, not a photo they upload from their camera roll. Several frames are captured; quality gating picks a usable one before matching runs.
How it works
Detect, align, compare, report.
Three steps, each of them a separate, reported piece of evidence.
Detect and align
Face detection and landmark alignment are built for the conditions a phone camera actually produces: off-angle heads, mixed lighting, motion blur and document photos that are years old or slightly worn.Compare
The two faces are compared directly against each other. Nothing is checked against a watchlist or a database of other people's faces here — that is a separate, opt-in add-on.Report
Face Match returns its own pass or fail, alongside — never instead of — the document, liveness, anti-spoof, deepfake, mask and tamper checks that run on the same verification.
How the result is reported
Pass or fail against a fixed line. Nothing in between, on your screen.
Under the hood, the model produces a similarity score. What you see is simpler: the check passed, or it didn't, measured against a decision line that does not move between verifications.
- One verdict per verification: pass or fail, reported the same way as every other check.
- The decision line is fixed. It is not tuned per customer and does not drift with volume or plan.
- A failed face match tells you which side the evidence points to, so the applicant can be told what to retake, not just that something went wrong.
Why no percentage
A lot of vendors print a number. We think that number is usually a guess.
A same-person probability like "99.7% match" only means something once it has been calibrated: measured against a large set of real genuine and impostor comparisons, on your own capture pipeline, on real phone photos. Skip that step and the number is a marketing decoration wearing a lab coat.
We haven't measured it on real traffic yet
We will publish a measured false-match rate, with the sample size it was measured on, once it is measured on real, consented verifications, not borrowed from someone else's dataset.
Until then, pass/fail is the honest answer
The decision line tells you whether the two faces are the same person for the purpose of this verification. That answer doesn't need a fake percentage to be useful.
A number with no sample size is not evidence
If a vendor shows you "99.9% accurate" with no stated sample size, no stated dataset and no stated condition, you are looking at a claim, not a measurement. Ask them for the sample size. Then ask us — we'll tell you we don't have one yet either, and why that's the more trustworthy answer.
Selfie quality
A bad photo gets a retake, not a false result.
Before Face Match ever runs, the live capture is scored on five independent qualities. A capture that fails gets sent back for a retake instead of being forced through a comparison it can't support.
- Sharpness
- Lighting
- Pose
- Face visibility
- Occlusion
If a verification does fail, retake guidance looks at the actual evidence — a clean selfie against a steeply angled, unreadable document photo, for example — and tells the applicant which side to retake. Only that side's camera reopens. Nobody is asked to redo a selfie that was never the problem.
The send-as-you-go fast path
The check starts the moment the selfie lands.
Most verification services wait for the whole session to finish, then work through checks one at a time. ShieldID starts Face Match, anti-spoof, deepfake and mask detection the instant a usable selfie frame arrives, often before the liveness challenge itself has finished. The applicant keeps completing the challenge on their phone while the biometric checks are already running in the background.
We don't publish a processing-time promise in seconds — no vendor's number survives contact with a bad phone signal — but starting early instead of queuing everything serially is the difference between a result that's ready when the applicant finishes and one they wait on.
What Face Match doesn't claim
One honest signal, not a fraud department in a box.
Face Match's job is narrow on purpose: confirm the live face matches the document portrait. Here is what that does, and doesn't, cover.
A genuine document, borrowed
If someone hands over a real, unaltered ID that belongs to a person who closely resembles them — a sibling, a parent, in some cases a twin — a face match alone may not catch it. That's why Face Match is one signal among several: document reading, liveness and per-field verdicts all run on the same verification, and none of them alone is the whole answer.
A synthetic identity
A fabricated identity built from a mix of real and invented information has no genuine "real person" on file to compare against. Face Match confirms the selfie matches the document in front of it — it cannot tell you whether that document's identity was invented in the first place. Watchlist and sanctions screening, run as an add-on, work a different angle on that problem.
Document authenticity
Face Match does not decide whether the document itself is genuine, tampered with, or a well-made forgery. That's a separate check, reported on its own, and it currently advises rather than gates the verdict.
Privacy
The photos do their job, then they go.
Face Match needs the document portrait and the live capture only for the moment of comparison.
- Images are encrypted at rest.
- ID images, including the selfie used for matching, are purged automatically on a short timer, seven days by default.
- Your applicants' photos are never used to train our models or anyone else's.
Questions
Straight answers about the match.
See it match your own face
Run a sandbox verification and watch the check report pass or fail.
No sales call, no percentage to argue with. Sandbox sessions are never billed.