Morphable Face Reconstruction With Multiple Cameras

The modeling of realistic looking human faces is one of the most interesting and dif-ficult problems in computer graphics and computer vision. Since the pioneering work of Parke, various modeling techniques have been adopted to reconstruct 3D human faces dur-ing the past few years.The 3D morphable face model uses the statistical method and shape from shading technique to recover 3D shape and texture of faces from a single image. It is based on prior knowledge of exemplar faces, thus the 3D morphable face model can yield excellent result between lowly textured face images and has become one of the most popular modeling methods of human faces. In many scenarios, it is very useful to be able to estimate 3D shape from only one single image. However, since one single image cannot provide depth information, and the estimation of depth is just based on the statistical correlation of 3D exemplar faces, the 3D shape cannot be estimated with great precision. For the solution of this problem, more images from different views are used to estimate the depth information, and the reconstructed 3D shape of the face would be more accurate.In this paper, an automatic and efficient method for realistic 3D face reconstruction based on multiple views is propoesed. At the beginning, a 3D morphable face model is es-tablished by computing the dense point-to-point correspondence between the reference face and exemplar faces. Then the morphable face model is matched to multiple images from different views at the same time to reconstruct given faces. We also employ a maximum a posteriori (MAP) estimator to enhance the robustness of the algorithm. With the informa-tion from multiple images, our method can obtain more realistic reconstructed effects. In addition, we construct a multi-camera system based on our algorithm for fully automatic 3D face reconstruction.