Abstract
This dissertation provides a methodology for processing and building 3D mesh models of human faces, targeting 3D to 3D and 2D to 2D face recognition applications. In particular, we developed two algorithms for face modeling from either stereo or range images and one algorithm based on 3D model face image synthesis. The first algorithm produces a deformed 3D model using stereo images from two parallel frontal views and one orthogonal profile view of the face, while the second algorithm produces a deformed 3D model from one range image of the face. Both algorithms obtain 3D models which are applied in 3D face recognition. Additionally, the stereo-image-based algorithm is capable of synthesizing multiple views 2D facial images which are applied in 2D face recognition.Unlike methods that use stereo vision alone or methods that use range images for face recognition, our stereo-based algorithm has the advantage of combining 3D information of depth from stereo with 2D extracted labels for facial feature points. Both stereo and range image approaches use a generic mesh model which establishes direct feature vertices correspondences between all subjects, resulting in a faster and more accurate face recognition comparison.