Abstract

Frame-Level 3D Biometric Provenance for Video describes a method for proving that a previously authenticated person is genuinely present in every frame of a video recording, closing a gap that existing content provenance and deepfake defenses leave open.

Signing a video proves the file has not been altered after capture. It does not prove that the person on screen is who they claim to be, or that their face was not replaced. This disclosure addresses that gap with three linked techniques:

  1. Trust handoff: an initial strong authentication of the subject (for example a passkey, ID check or human verifier) is cryptographically bound to a 3D facial reference, without storing raw biometric data.
  2. Frame-by-frame 3D verification: a calibrated stereo or depth camera, such as a VR180 rig used as a companion to a primary camera, verifies the subject's 3D facial geometry in each frame. Cross-view geometric consistency serves as an anti-synthesis check, since a real face must agree across two calibrated views and over time.
  3. Signed per-frame Merkle tree: each frame's image data and verification result are hashed into a Merkle tree whose root is signed in secure camera hardware. Any frame or excerpt can be independently verified, and tampering is localized to the affected frames.

The disclosure also covers:

  • explicit status codes for gaps, occlusion and multiple subjects
  • segment-based signing for live streaming
  • handling of re-encoding through perceptual hashes and watermarks
  • privacy-preserving subject tokens
  • a range of alternative capture hardware and cryptographic embodiments

The method is compatible with the C2PA Content Credentials framework. It is published as an open defensive disclosure, free for public use.

Creative Commons License

Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.

Download

Share

COinS