Abstract:

To improve the accuracy of 3D face reconstruction, a novel coarse-to-fine 3D reconstruction method is proposed. With salient points in each region, a rough shape is reconstructed based on SRSD (Surface Reconstruction from Sparse Data) method. Since the shape reconstructed is a rough approximation of real model, the following reconstruction based on this shape is more accurate. Taking the rough shape as reference model, distance of feature points between 2D face and the projection of reference model can be calculated. Then deforming factors are computed and the final precise shape is reconstructed. Since the roles of each reconstruction step are different, parameters in each step are selected in different ways. Experimental results show that this method can effectively improve the accuracy while having similar computation speed.

Key words: 3D face, morphable model, sparse reconstruction, multi-step reconstruction