Face match compares a selfie to the photo on an ID and returns a similarity score against a threshold, to confirm the same person is present. It is a core defense against impostors using someone else's document, but it only means something with liveness attached.
What is face match, in plain English?
Face match takes two images, the selfie a user submits and the photo on their ID, and measures how similar the faces are. It returns a similarity score, and you compare that score to a threshold to decide pass or fail. The goal is simple: confirm the person presenting the document is the person the document describes.
It is a core control against impostors, because a stolen but genuine ID gets you nowhere if your face does not match the photo. But a score alone is fragile. A fraudster can drive a high match with a printed photo, a mask, or a deepfake of the real owner's face. That is why face match only means something with liveness attached, confirming the selfie came from a real, present human rather than a replay.
A specific threat to watch is the morphing attack, where two faces are blended so one document photo matches two different people. In fraud and AML, face match is the bridge between the document check and the human, but it is one signal in a stack, not a verdict on its own.
Setting the threshold
What changes | Threshold set high | Threshold set low |
False accepts | Fewer impostors slip through. | More impostors can pass. |
False rejects | More genuine users get bounced. | Fewer good users get bounced. |
Customer friction | Higher, more retries and reviews. | Lower, smoother onboarding. |
Best fit | High-risk or high-value flows. | Low-risk flows where friction hurts. |
Who is involved?
Who | Their role |
The applicant | Submits a selfie to be compared against the ID photo. |
The biometric vendor | Computes the similarity score and returns it against the configured threshold. |
The risk owner | Sets the threshold to balance false accepts and false rejects for each flow. |
The impostor | Presents a stolen ID and tries to beat the match with a photo, mask, deepfake, or morph. |
What it looks like in practice
In practice
An applicant submits a genuine passport and a selfie. The face match scores high and, on a face-match-only flow, would pass instantly. But the liveness check flags that the selfie has no depth and reflects a flat surface, consistent with a photo held to the camera.
The fraudster had a real stolen passport and a clear photo of the owner, enough to beat the match but not the liveness step. On a separate case, a document photo quietly matches two different applicants; that turns out to be a morphed image built to serve both. Face match found the similarity in both; liveness and morph checks decided what it meant.
Why it matters to operators
Face match is what stops a stolen document from being enough. It ties the paper to a person, which closes the gap document authentication leaves open. For most digital onboarding, it is the single most important biometric signal you run.
The discipline is remembering what the score is not. Without liveness, a high match is spoofable with a photo, mask, or deepfake, and a morphing attack can make one document pass for two people. Set the threshold against how much false accept and false reject you can tolerate for each flow, and always read the score alongside liveness and the document check rather than alone.
What to watch in the data
- High match, no liveness. A strong score with a weak or missing liveness signal is exactly how photo and mask spoofs pass.
- One photo, two people. A document image that matches more than one applicant points to a morphing attack.
- Score clustering at threshold. Many cases landing right at the cutoff can mean someone is tuning attacks to just clear it.
- Lighting and depth mismatch. Flat, evenly lit selfies with no depth cues suggest a screen or printed replay.
- Repeated selfies, different IDs. The same face matched against several documents may signal a mule or synthetic operation.
Quick questions
Is a high face match score proof of identity?
No. It proves two images look alike. Without liveness, that similarity can come from a photo, mask, or deepfake, so a high score only carries weight when a liveness check confirms a real, present human.
How do I set the threshold?
Balance false accepts against false rejects for the specific flow. High-risk actions warrant a stricter threshold even at the cost of more genuine bounces; low-risk flows can loosen it to reduce friction.
What is a morphing attack?
It blends two or more faces into one document photo so the ID matches multiple people. An ordinary face match can pass a morph, so detecting it needs morph-specific analysis or comparing the chip photo to the printed one.
How is face match different from facial recognition?
Face match is one-to-one: it compares a selfie to a specific ID photo. Facial recognition can also be one-to-many, searching a face across a gallery to spot, for example, a repeat fraudster.
Why do genuine users sometimes fail?
Poor lighting, pose, aging since the ID photo, or a low-quality document image can push a real match below threshold. Building a retry or manual review path avoids rejecting good customers on a single low score.
Go deeper
- NIST Digital Identity Guidelines (SP 800-63) ↗ — The US standard for identity proofing and authentication assurance levels.
- FATF ↗ — The global standard-setter for AML, counter-terrorist-financing, and counter-proliferation. Recommendations, guidance, and jurisdiction lists.