How AI face matching works

You upload one photo, and a few seconds later you get a ranked list of lookalikes with a confidence score for each. Here's what actually happens in between — no maths degree required.

At its heart, face matching is about turning something messy and human — a face — into something a computer can compare precisely: a list of numbers. Dollganger does this in four steps.

1. DetectionFirst the system finds the face in your photo and ignores everything else — background, clothing, other people. Dollganger uses a detector called RetinaFace, which also locates key points like the eyes, nose and mouth corners.
2. AlignmentUsing those key points, the face is rotated and cropped to a standard position. This way a head tilted slightly to the left is compared fairly against one looking straight ahead.
3. EmbeddingThe aligned face is fed through a neural network (ArcFace) that outputs a 512-number "face signature". Faces that look alike end up with similar signatures, even across different lighting and expressions.
4. ComparisonYour signature is compared against every face in the database using cosine similarity — a measure of how closely two lists of numbers point in the same direction. The closest matches become your results.

What does the confidence score mean?

The percentage next to each match is that cosine similarity, expressed in a friendly way. A high score means the two face signatures are very close; a lower score means the resemblance is looser. It is not a statement that two people are related — just that the geometry of their faces lines up well.

Why 512 numbers?

Each of those numbers captures some abstract aspect of facial structure the network learned from millions of examples — things like the ratio between features, the shape of the jaw, the set of the eyes. No single number means "nose width"; the information is spread across all 512. Together they're enough to tell millions of faces apart while still grouping lookalikes close together.

Where your photo goes

Nowhere permanent. Your image is processed in memory to compute its signature and is then discarded — it is never saved to disk or to a database. We cover this in detail in What happens to your photo.