An accurate and efficient method for reconstruction of 3D faces from stereo images

In this thesis, we introduce a novel algorithm for reconstructing the 3D shape and texture model of human faces from two stereo images which are captured from calibrated cameras. Our approach works in a sparse to dense manner: we first build a coarse shape estimation based on 3D keypoints, and then use a linear morphable model to efficiently match the detailed shape and texture. The features used for the fitting processes are selected with the guidance of the quantitative evaluation of a state-of-the-art reconstruction algorithm. In our new direct evaluation method, the reconstructed 3D faces are first aligned to the ground truth and then the shape difference between the two 3D faces is described by signal-to-noise ratio and error maps illustrating the reconstruction errors on corresponding vertices. This local error information will be used to resample the reference frame whose vertices' coordinates stack up to be the feature vectors for face fitting. Compared with the previous works, our algorithm can reconstruct the 3D face shape at a speed comparable with that of the fastest algorithm available, but gives a higher accuracy. It can also recover the more complete and realistic looking texture. Our results show that the new algorithm possesses significant characteristics of a 3D face model reconstruction system, and is especially useful for face recognition and animation applications in practice