Geometric and photometric head modelingfor facial analysis technologies

We present geometric and photometric head modeling techniques that are
computationally efficient, and yet achieve a high level of realistic animation.
First, we reconstruct a textured face model from range data obtained by a
Cyberware scanner. The geometric model is selectively re ned at features of
interest while simultaneously extrapolating missing data, and the final head
model is suitable for real-time facial animation. We then propose a photometry
compensation algorithm using the OpenGL graphics library, that reduces the
photometric discrepancies at the 3D level between a synthesized view of the
model and the same view of the real person. We evaluate the performance of
the proposed algorithm using computer generated images.
Due to the realism of the geometric and photometric models, enhanced
analysis/synthesis cooperations are made possible in such applications as face
cloning, head tracking, face recognition, video indexing or person authentication.