Learning a generic 3D face model from 2D image databases using incremental Structure-from-Motion

Over the last decade 3D face models have been extensively used in many applications such as face recognition, facial animation and facial expression analysis. 3D Morphable Models (MMs) have become a popular tool to build and fit 3D face models to images. Critical to the success of MMs is the ability to learn a generic 3D face model. Towards that end, major limitations in the current MMs’ process are: (1) collecting 3D data usually involves the use of expensive laser scans and complex capture setups, (2) the number of available 3D databases is limited, and typically there is a lack of expression variability, (3) finding correspondences and registering the 3D model is a labor intensive and error prone process. This paper proposes an incremental Structure-from-Motion (SfM) approach to learn a generic 3D face model from large collections of existing 2D handlabeled images, containing many subjects under different expressions and poses. Two major contributions are: (1) learning a generic 3D deformable face model from 2D databases, (2) incorporating a prior subspace into the incremental SfM formulation to provide robustness to noise, missing data and degenerate shape configurations. Experimental results on the CMU-PIE database show improvements in the generalization of the 3D face model across expression and identity.