Face recognition and computer graphics for modelling expressive faces in 3D

Abstract

This thesis addresses the problem of the lack of verisimilitude in animation. Since computer vision has been aimed at creating photo-realistic representations of environments and face recognition creates replicas of faces for recognition purposes, we research face recognition techniques to produce photo-realistic models of expressive faces that could be further developed and applied in animation. We use two methods that are commonly used in face recognition to gather information about the subject: 3D scanners and multiple 2D images. For the latter method, Maya is used for modeling. Both methods produced accurate 3D models for a neutral face, but Maya allowed us to manually build 3D models and was therefore more successful in creating exaggerated facial expressions.