Face Quality & Match Checker

Score a face photo against the ISO/IEC 29794-5 (OFIQ) standard, and check whether two photos are the same person — using the same engine behind our enrolment and verification platforms.

Compare two faces

Both photos are processed in memory and discarded the moment you have your answer. Nothing is saved.

Photo A

Photo B

A quick check, so the free tool stays available to everyone.

Or try a sample — these don’t use your daily runs

Result

A Match Level against a fixed decision threshold, plus an ISO/IEC 29794-5 quality assessment of each photo.

Choose two photos, or run a sample, and the verdict appears here.

What OFIQ and ISO/IEC 29794-5 actually measure

A face photograph can look perfectly good to a person and still be a poor biometric sample. ISO/IEC 29794-5 exists to make that judgement measurable rather than a matter of opinion: it defines a set of quality components, each scored 0–100, and a single unified score predicting how well the image will perform in a biometric comparison.

OFIQ — Open Source Face Image Quality — is the German Federal Office for Information Security’s reference implementation of that standard. Because it is the reference implementation, a score from OFIQ means the same thing to any other party using it, which is why it has become the common currency for face image quality in passport, visa and national identity programmes.

Some of the 28 measures, and what causes a low score
Measure	What it looks at	What drags it down
Sharpness	Focus and motion	Camera shake or a subject who moved during capture.
Illumination	Lighting uniformity	Light from one side, leaving one half of the face in shadow.
Head pose	Yaw, pitch and roll	The head turned, tilted or nodded away from square to the camera.
Occlusion	Coverage of the face	Hair, a hand, a mask or heavy frames obscuring facial features.
Expression	Neutrality	A broad smile or open mouth, which distorts the geometry a matcher relies on.
Background	Uniformity	A busy or patterned background behind the subject.
Eyes	Visibility and gaze	Closed or partly closed eyes, glare on glasses, or gaze off-camera.

Every measure runs the same direction: higher is always better. That is worth stating because several of them are named for the defect they detect — a high “compression artefacts” score means few artefacts, not many.

How the match decision is made

Face matching does not compare pictures. Each face is converted into a template — a fixed-length list of numbers describing the geometry and texture of that face. Comparing two templates produces a Match Level on a fixed ML 1–7 ladder, each level a named operating point with its own known false-match rate.

The operating point is the part that matters operationally, and it is a policy decision rather than a technical one: how often you are willing to accept two different people being called the same person. This tool runs at roughly one false match in ten thousand comparisons, a reasonable general purpose setting. A border control system would run tighter; a photo-library grouping feature would run looser.

This is also why a borderline result (low ML, close to the decision threshold) is worth treating differently from a strong one — a well-designed production system routes a borderline match to a human rather than deciding on its own.

What happens to your photos

They are held in memory for the seconds the comparison takes, and then they are gone. Nothing is written to disk, no copy is retained, and no template is enrolled into any database. There is no version of this page that keeps your face.

What we do keep is the result: the match level, the quality outcome for each image, and your rating if you leave one. That is what tells us whether the tool is behaving sensibly. Faces are special-category personal data under both POPIA and the GDPR, and a free public demonstration is not a good reason to hold anyone’s.

Frequently asked questions

What is OFIQ?

OFIQ (Open Source Face Image Quality) is the German BSI’s reference implementation of ISO/IEC 29794-5, the international standard for measuring how usable a face photograph is for biometric matching. It returns 28 standardised measures — sharpness, illumination uniformity, head pose, expression, occlusion and more — plus a single unified quality score from 0 to 100.

What does the quality score actually tell me?

It predicts how well that image will perform in a biometric comparison — not whether it looks nice. A photo can be attractive and still score badly for enrolment: strong side lighting, a slight head turn, glasses reflections or a busy background all reduce how reliably a matcher can use it. That is why passport and national ID programmes gate enrolment on quality before a photo enters the register.

How does face matching decide "same person"?

Each face is converted into a mathematical template — a list of numbers describing the face, not a stored picture. Comparing two templates produces a Match Level, checked against a threshold. The threshold is chosen from a target false-match rate: this tool uses one calibrated to roughly one false match in ten thousand comparisons.

Why is a match not 100%?

A match is never reported as a percentage, and it is not meant to be — the comparison produces a Match Level, not a confidence score. A high match level (ML 6-7) already means the result sits far clear of where two different people ever land, an emphatic match rather than a marginal one. Two genuine photographs of the same person differ in lighting, pose, expression, camera and age, so even a strong match is never a perfect, identical comparison — that is expected, not a weakness. Comparing photos taken in very different conditions — a phone selfie against a formal enrolment portrait, say — will generally land at a lower match level than two studio photographs, without meaning the match is any less certain.

What is a "match level"?

It is the same ML 1–7 ladder RealFace uses internally, and it exists for the same reason a security researcher would rather hear "AES-256" than a description of the math: each level is a fixed operating point with a known false-match rate, from roughly 1 in 100 at ML 1 up to 1 in 500,000 at ML 7. "ML 7" is a decision you can act on — the higher the level, the harder that result would be to fake by chance.

Are my photos stored?

No. Images are held in memory only for as long as the comparison takes, then discarded. Nothing is written to disk, and no copy is kept. We record the resulting numbers — the match score and quality outcomes — so we can see how the tool is performing, along with your rating if you leave one.

Why is there a daily limit?

Face matching and quality assessment are computationally expensive, and this is a free public demonstration running on shared hardware. The sample photos are exempt from the limit, so you can always see how the tool behaves. If you need it for real work, talk to us about API access.

Can I use this in production?

Not this page — it is a demonstration with fixed settings. The engine behind it is production software and is available as an API, deployable in your own environment, with thresholds calibrated to your own risk appetite and the quality policy tuned to your programme’s rules.