Notes
🔍

Face Recognition

Face recognition lets a computer determine whether faces belong to the same person, or find a known person within a collection of identities. Rather than “seeing” a face as a human does, the system converts visual patterns—such as the arrangement of eyes, nose, mouth, and facial contours—into numbers it can compare.

From face image to identity comparison

A practical system begins with face detection, which locates faces in an image or video frame. It then estimates landmarks and aligns the face: rotating and scaling it so the eyes and other features occupy consistent positions. A neural network turns the prepared face into a compact embedding, a vector of numbers representing identity-related features. Faces from the same person should produce nearby vectors; faces from different people should be farther apart.

Two common decisions
  • Verification (one-to-one): “Is this passport photo the same person as this camera image?” The system compares two embeddings and accepts or rejects the match using a threshold.
  • Identification (one-to-many): “Whose face is this?” The embedding is searched against a gallery of enrolled people, returning the closest candidates.
Why accuracy needs careful design

Models such as FaceNet learn embeddings by training on many labeled faces and explicitly rewarding useful distances between identities. This supports phone unlocking, photo organization, access control, and locating a missing person in authorized footage. But the comparison threshold is crucial: too loose produces false matches, while too strict rejects legitimate users. Real systems must also handle pose, poor lighting, aging, masks, image quality, demographic performance differences, and presentation attacks such as a printed photo. Liveness detection helps distinguish a live person from a spoof. Because a face is sensitive biometric data, responsible deployment also requires consent, security, limited retention, and clear rules about where recognition is appropriate.

Face recognition is the task of identifying or verifying a person from facial images or video by comparing learned facial features against known identities. It distinguishes individuals despite variations in pose, lighting, expression, and image quality. Face recognition enables identity-based applications such as device authentication, access control, photo organization, and forensic search; its accuracy and bias directly affect security, privacy, and fairness.

Think of how you recognize a friend in a crowded café: you do not need to read their name tag. Their face gives you enough clues to know who they are. Face recognition gives computers a similar ability.

It is used to decide whether a face in a photo or video matches a known person, such as unlocking a phone, organizing family photos, or helping verify someone’s identity at an airport. It is different from simply noticing that a face is present; it tries to answer, “Whose face is this?” Because faces are personal information, face recognition can be useful but also raises important privacy and fairness concerns.