OCR will read a forged ID card as accurately as a genuine one. Reading and trusting are different jobs.
Extraction answers what does this card say? A forger's whole aim is for that answer to come out clean. So an eKYC system that is excellent at extraction and has nothing else is, against a competent forgery, an efficient way of onboarding the wrong person. Document fraud needs its own checks, looking for things extraction is not built to notice.
Legible is not the same as genuine
Five signals extraction misses
| Signal | Example | Why extraction alone misses it |
|---|---|---|
| Forged document | A card produced from scratch | Every field is printed to be read cleanly |
| Tampering | A photograph swapped, a number or date edited | Each altered field is still perfectly legible |
| Blacklist match | A document or identity already known to be fraudulent | The card itself may be entirely genuine |
| Anomaly | Fields that do not agree with each other, or with how the card should be laid out | Each field is valid on its own; only the combination is wrong |
| Suspicious behaviour | The same ID attempted repeatedly, or many attempts from one device | It is invisible inside a single session |
Expiry sits alongside these. A genuine card that has lapsed is not fraud, but it is not acceptable evidence either, and validating expiry at capture is cheaper than discovering it later.
The attack that beats a face match
One kind of tampering deserves particular attention. If an attacker replaces the photograph on a genuine card with their own, then the face match will pass — the selfie really does match the card. The liveness check will pass too, because a real person is present.
Only document tampering detection catches it. This is why the checks cannot be traded against each other: a strong face match is no reassurance about the card the face was matched to. It is the mirror image of the attack that liveness detection exists to stop.
Evidence, not verdicts
A fraud signal is a reason to look, not proof. Cards wear, laminate lifts, glare hides a corner, and a cheap phone camera adds noise that can resemble editing. A system that rejects on every irregularity will turn away genuine customers; one that hides its reasons will leave reviewers guessing.
The better shape is graded — proceed, review, reject, escalate — with each signal shown so that an authorised person makes the final call. It is the same principle as in detecting cheque alterations: the system prioritises evidence, and people decide.
Built for the cards you actually receive
Authenticity checks only work when they know what a genuine card looks like. ekyciq's extraction and authenticity checks are built for the Omani Resident ID and National ID, and its fraud detection looks for forged documents, tampering, blacklist matches, anomalies and suspicious behaviour.
See how ekyciq works, or talk to us about the documents your onboarding sees most.
Related posts
-
Face Match Is Not Enough: Why eKYC Needs Liveness DetectionA face match asks whether two faces are the same. Liveness asks whether anyone is actually there.
11 September 2026 -
On-Premise or Cloud Video Analytics: What Actually ChangesFootage that never leaves the site removes a compliance question rather than answering it.
05 May 2026 -
The False Alarm Budget: Why Noise Is the Real Failure ModeA system nobody trusts is worse than no system: you are paying for coverage you have already lost.
02 December 2025