What is Facial Recognition?
Face Matching Technology
📖 Definition
Facial recognition in KYC is the process of comparing a live selfie against the photo on your government ID to confirm both images show the same person. A system converts each face into a numerical template, measures how closely the two templates match, and returns a similarity score that decides whether your identity passes, fails, or needs manual review.
Under the hood, this happens in a few distinct steps. First, the system detects and isolates the face in each image, separating it from the background, clothing, or the printed border of an ID card. Next, it maps facial landmarks, the distance between your eyes, the width of your nose, the shape of your jawline, and the depth of your eye sockets, and normalizes the image for lighting and angle. A deep learning model then converts that landmark data into a faceprint, a compact vector typically made of 128 to 512 numerical values that represents the geometry of your face mathematically rather than as a picture.
To decide if you are a match, the system compares the faceprint from your selfie to the faceprint extracted from your ID photo using a similarity metric such as cosine similarity or Euclidean distance. That comparison produces a single score, and if the score clears a threshold set by the platform, the system records a match. If it falls short, you get a rejection, sometimes paired with a vague error message and sometimes with a specific code depending on the platform's KYC provider.
📊 Why Facial Recognition Matches Fail (Even For Real People)
A failed match is not proof of fraud. Because the system is comparing two images mathematically, ordinary differences between your selfie and your ID photo can push the similarity score below threshold even when you are exactly who you say you are.
| Cause | Why It Happens | How To Avoid It |
|---|---|---|
| Poor lighting | Shadows, glare, or dim rooms obscure facial features and distort contrast, making the extracted faceprint less reliable. | Face a window or light source directly, avoid backlighting, and skip a flash that causes glare. |
| Extreme angle | ID photos are taken near frontal, so a selfie shot from above, below, or in profile changes the visible geometry of your face. | Hold the camera at eye level and look straight into the lens for a neutral, front facing shot. |
| Outdated ID photo | A passport or license photo taken many years ago may no longer closely resemble your current face, increasing mismatch risk from natural aging. | Renew an ID that is very old where possible, and expect a stricter platform to route older documents to manual review. |
| Glasses or accessories | Thick frames, tinted or reflective lenses, hats, or masks cover or distort the eye and nose region the system relies on most heavily. | Remove glasses, hats, and face coverings for the selfie unless your ID photo shows the same item. |
| Low camera resolution | Blur, compression artifacts, or a low resolution capture reduce the detail available for feature extraction, weakening the faceprint. | Use a modern phone camera in good light and avoid re-compressing or screenshotting the photo before upload. |
🎭 Facial Recognition vs. Liveness Check
People often use these two terms interchangeably, but they answer different questions. Facial recognition asks whether two face images, your selfie and your ID photo, belong to the same person. A liveness check asks a separate question entirely: is there a real, physically present human in front of the camera right now, rather than a printed photo, a video replay, or a mask. Most KYC flows run both checks, one after the other, which is why it is possible to pass the liveness check by clearly being a live person and still fail the facial recognition match because your face does not resemble your ID photo closely enough, or the reverse.
🏢 Where You Will See This
Binance, Coinbase, Revolut, PayPal, and most banking and crypto KYC providers run facial recognition as a standard step, usually right after document upload and alongside a liveness check.
Frequently Asked Questions
Why does facial recognition reject real people?
Facial recognition compares two images mathematically, so it can reject a genuine user when the images differ too much for reasons that have nothing to do with identity, such as poor lighting, an extreme camera angle, an outdated ID photo, or heavy compression on the uploaded file. The system is not judging whether you are a real person, only whether the two faceprints are similar enough to clear its threshold, and that threshold has no way to know the difference between fraud and a bad photo.
Is facial recognition the same as a liveness check?
No. Facial recognition asks whether two face images belong to the same person, while a liveness check asks whether the person in front of the camera right now is a real, physically present human rather than a photo, video, or mask. Most KYC flows run both, usually one after the other, and it is entirely possible to pass one and fail the other.
Does facial recognition raise privacy concerns?
Yes. Facial recognition relies on biometric templates, which are treated as sensitive personal data under regulations like GDPR because, unlike a password, a face cannot be reset if that data is ever exposed. Researchers have also documented that some facial recognition systems perform less accurately for certain demographic groups, which can mean more manual reviews or false rejections for those users.
Can glasses or makeup cause a facial recognition failure?
Yes, particularly thick frames, tinted lenses, or glare across the lenses, since these obscure the eye region that facial recognition relies on heavily. Heavy makeup, dramatic contouring, or anything that changes the visible geometry of the face can have a smaller but real effect, and the safest approach is a plain, well lit photo with your face fully visible and no accessories that were not in your original ID photo.
Having KYC Issues?
Try our free AI analysis or get expert help for guaranteed approval.