Reconstruction of 3D Face Model from Single Shading Image Based on Anatomical Database

We propose a method to reconstruct 3D face model from a single face-frontal shading image based on a database of the original 3D face shapes which are created according to anatomical basis of the human head. In the proposed method, we reconstruct the 3D shape of the face from the input shading image by estimating small number of eigenvalues taken by principal component analysis (PCA) of the anatomical database of the head, rather than directly recovering the shape from the input shading image. The eigenvalues are estimated based on the optimization of the error value computed by comparing shading information of input image to that of model image created from the shape represented by the eigenvalues. For evaluating the effectiveness, we reconstruct various face shapes from single shading images of the objects based on the proposed method. The reconstructed shape for a statue of head with the average shape of the databases provides an error evaluation, which is 2.6 mm in average. This is sufficiently accurate for the future applications, such as on-line order-made of face-wearing products