Fast model-based 3D human face reconstruction from silhouettes of multiple views

Recovering the 3D face model of a human head from photographs has been a topic of research in computer graphics for years. In this paper we present a novel model-based approach which solely depends on silhouettes of the input object. A PCA face model derived from USF HumanID dataset is used to model the 3D face. Instead of considering the entire silhouette of the head image, we only fit the model to the face outline of the silhouette which can eliminate the fitting error caused by the presence of hair. Parameters of the PCA model are adjusted according to the distances between the projected boundary vertices of the model and the silhouette. Our algorithm produce satisfactory model and is computational simple and efficient. Working on a high resolution model requires less than a minute to converge.