Abstract
In this thesis, we propose a novel approach to 3D hair modeling from multi-view images of a specific person. The reconstructed 3D hair model can be integrated with the reconstructed 3D face model to obtain a complete human head model for vivid facial animation. In our 3D hair reconstruction system, we first estimate 2D orientation of the hair regions in each view, and estimate an initial 3D orientation field of hair according to the relative position to the reconstructed face model. Then the initial 3D orientations are refined by optimizing a cost function carefully designed to combine multi-view constraint, 3D orientation constraint, and strand constraint. Then, our algorithm grows the 3D hair strands from the root points, which are uniformly sampled on the pre-defined surface region of the reconstructed 3D head model, along the direction determined by the refined 3D orientation field. Finally, we show some experimental results of the reconstructed 3D hair strand models by using the proposed algorithm.