Abstract: An algorithm is proposed to generate 3D face model by depth map with a pair of stereo images. By the algorithm,a pair of images is first captured,and rectified to align the epipolar lines and compensate for the image distortions.Then image matching is accomplished based on seed voxels where a novel seed voxel selection method is proposed.The method guarantees that sufficient seed voxels be selected,with accurate disparities among the seed voxels.Besides,a disparity expansion algorithm based on the reliability of the disparity is also proposed,and it minimizes the area where erroneous disparities occur. After image matching, the 3D face model is reconstructed,and the face surface is smoothened with particle-based representation and Delaunay triangulation.Experimental results show that the reconstruction results are smooth and vivid.