13.4 Facial Recognition

Facial recognition turns a face into a compact numeric vector called an embedding, where similar faces sit close together. Two tasks are common: verification (is this the same person as on the ID? one-to-one) and identification (who is this among many enrolled people? one-to-many). Matching is done by measuring the distance between embeddings.

A bank branch might use one-to-one verification to confirm a customer matches their stored photo before a large withdrawal. The system only compares two faces, not search a national database, which keeps it focused and lower-risk.