B-Lab: Biometrics and security

2D and 3D face analysis for unconstrained face recognition in the wild.

Face recognition facility

Impact: World-leading face recognition technology

  • Spin-out Omniperception (face verification)
  • European biometrics industry achievement award (2017)
  • Face recognition beats human super-recognisers (2018)

Current activities

  • 3D morphable face model building from 3D face images
  • 3D assisted 2D face verification and recognition
  • 3D face reconstruction from video
  • Deep neural networks for face matching
  • Person re-identification
  • Anomaly detection for face anti-spoofing
  • Linking 3D face shape to genetic markers

Current projects

Distinctive approach and achievements

  • Using 3D face model:
  1. As a prior in machine learning
  2. For training-data augmentation by face image synthesis
  3. 3D face reconstruction from video.
  • Fusion of multimodal sources of information
  • Accelerated 3D face model to 2D image fitting to facilitate real-time face processing by a factor of 30
  • Pioneered multi-cohort 3D morphable face model
  • Single gene variants uncovered (Proc. Nat. Academy of Sciences 2018).

Future focus

  • Modelling image degradations (low resolution, blur, occlusion)
  • Image quality controlled face analysis
  • Joint image and language re-identification
  • Face spoofing detection formulated as anomaly detection
  • Developing information theory for decision-making applications