New Method for Locating the 3D Facial Landmark Points

Locating landmark points plays an important role in applications related to human faces, such as registration of 3D faces, the deformation based on the fiducial points, and so on. We present a new form of three-dimensional shape representation based on the local projecting area, and suggest an algorithm of locating the landmark points with it. This representation is invariant to rotation and translation, and need no training of gallery face set. Comparing with the classical method such as that using shape index, it doesn´t require complex curvature computation of point cloud, without demand of large local neighborhood and the high quality of point cloud. So the proposed algorithm has lower amount of computation and better veracity.