Computational Anatomy for Generating 3D Avatars and Boosting Face Recognition Systems

In this paper we present results of an automated system for generating 3D avatars from one or more photographs. We are motivated by the need to support pose and lighting correction for invariant facial identification. Our approach is to extend techniques from the now well established Computational Anatomy field to accommodate the projective geometry associated with video imagery. We present the general Compuational Anatomy framework, describe its merger with and application to the projective geometry setting, and present validation results on FRGC EXP 1.0.1, FRGC EXP 1.0.4, and FERET databases. In particular, we find sub-degree bias in rigid motion accuracy for avatar generation, and 1/20 eye distance average errors in feature landmark accuracy.