Synthesis of 3d Faces

In this paper, we present a new technique for creating photorealistic textured 3D facial models from photographs. Starting with several pairs of calibrated views of a human subject, we first automatically recover a 3D dense reconstruction for each pair. We employ a user-assisted technique to align these 3D meshes with a generic face mesh. A scattered data interpolation technique is then used to deform the generic mesh to fit the particular geometry of the subject’s face. Having recovered the geometry we automatically blend the different texture maps to obtain a final 3D face with a high degree of realism. The rich texture map is 512x1024 pixels and captures the complete head. Faces produced by our method are only 1700 polygons and suited to realistic animations of specific individuals in applications like special effects, games, and 3D teleconferencing.