3D Face Recognition Algorithm

Overview of the 3D algorithm

1. Preprocessing: Reconstruct a 3D mesh of the face from the 3D camera

Raw 3D Data

2. Model Fitting: Fit a parametric, annotated face model (AFM) to the mesh using a deformable model framework with additional anthropometric constraints

Parametrized, Annotated Face Model
Fitted AFM

3. Parametric Deformation Image: Use the parameterization of the fitted AFM to generate a three-channel deformation image of the face by encoding its geometry. The XYZ displacement is represented as RGB colors in the image below.

Deformation Image

4. Metadata Extraction: Compress the deformation image using a wavelet transform