Liveness detection

Confirming a real, present human is in front of the camera, not a photo or a video.

When a person takes a selfie for verification, liveness detection checks that the face belongs to someone actually there, instead of a picture held up to the lens or a recording played back.

Determination that a biometric sample originates from a live subject rather than an artifact or replayed signal. An artifact is a fake such as a printed photo, a mask, or a screen showing a face.

Liveness systems look for cues that a live face produces and a fake does not:

• Texture and detail, such as skin patterns that a print or screen flattens.
• Depth and shape, which show a three-dimensional face rather than a flat image.
• Reflections and lighting, which behave differently on skin than on glass or paper.
• Motion, such as tiny natural movements or a response to a prompt.

Systems use one of two approaches:

• Active liveness asks the person to perform an action, such as blinking or turning their head.
• Passive liveness checks the selfie for these cues with no action from the person.

Liveness detection sits next to terms that are easy to mix up:

• Presentation attack detection (PAD) is the formal, standardized name for detecting fakes held up to the camera, and liveness detection is the term most products use.
• Personhood is the broader goal of confirming a live human is present, which also depends on defenses against injected video.
• Face match (1:1) compares a face with a reference photo, while liveness confirms the face is real.