Morphing attack

Blending two faces into one photo that matches both people.

For example, two people might combine their photos into one image that looks like both of them, then use it on an ID document so that either person can pass a face check against it.

Generation of a composite facial image that satisfies match thresholds for multiple enrolled identities. A match threshold is the similarity score a face comparison must reach to count as a match.

A morphing attack typically works like this:

1. Two people supply photos.
2. Software blends the two faces into a single image that resembles both.
3. The blended photo is submitted for an ID document or an enrollment.
4. Either person can later pass a face comparison against that photo.

Morphing attacks usually target the moment a photo is submitted, such as a document application or an enrollment, rather than the live capture.

Detection falls into two approaches:

• Single-image detection examines the submitted photo alone for signs of blending.
• Differential detection compares the submitted photo with a trusted live capture of the person to see whether they agree.

Morphing attacks sit next to terms that are easy to mix up:

• Photo substitution replaces the portrait on a document with a different person's face, while a morph blends two faces into one.
• Face-swap deepfake places one face onto another in a video, while a morph is a still-image blend.
• Presentation attack detection catches fakes shown to the camera, and a morph is submitted earlier as a document photo.