3D Eigenfaces for Face Modeling
Takuma Iwasa, Takuya Shima, M. Yashwanth Sai, Gang Xu · 2002
In this paper we propose the idea of approximating an arbi-trary 3D face model by a linear combination of ”3D Eigen-faces”. First, a set of 3D face models are constructed by a face modeling system and the principal component analysis is performed over the data set. The eigen vectors associated with the largest eigen values are extracted to form the eigen space. We call these eigen vectors ”3D Eigenfaces”. Any face model can then be approximated by a linear combina-tion of these eigenfaces, thus reducing the whole space of face shape to the smaller space of these coefficients. We show that the reconstructed faces using these 3D eigenfaces are sufficiently close to the original faces. This result can be applied in image-based reconstruction of 3D face models. 1