Three proofs, one decision
Trust is earned in order.
A face match by itself only says two images look alike. Liveness is not a gesture animation, and face match is not a cloud lookup—both are our own models, running inside the phone’s trust boundary. No proof is produced until the credential, the live sample and the link between them all hold.
- 01
The credential is authentic.
The phone validates the credential’s digital signature against a trusted issuer key. A photograph of an ID cannot supply that chain of trust.
- 02
Liveness model The person is live.
Ollet’s own model evaluates the live camera stream for evidence of a real person at the device, stopping printed photos, screen replays and other presentation attacks before they reach the matching step.
- Input
- Live camera frames
- Decision
- Live / not live
- Role
- Guards the sample
- 03
Face model The live person is the holder.
A separate model compares the live face one-to-one with the issuer-protected portrait, measuring similarity against a decision threshold. It asks “same person?”—never “who is this?”, so there is no gallery to search.
- Input
- Live face + credential portrait
- Decision
- Match / no match
- Role
- Binds person to credential