三维人脸姿态校正算法研究

Study on 3D face pose recovery algorithm

针对目前三维人脸姿态校正算法的鲁棒性差和计算复杂的问题,提出了一种基于法向量对齐的人脸姿态校正算法,即先通过计算离散高斯曲率获取面部的3个特征点,然后计算这3个特征点组成的平面的法线方向,最后通过该法线的朝向进行人脸姿态校正。该方法计算量小,适用性广,实验结果表明,校正效果好。

Many existing 3D face pose recovery algorithms have unsatisfactory robustness and high calculation complexity. In this paper, a normal vector alignment method is proposed for 3D face pose recovery. The proposed method detects three characteristic points on a human face through calculating discrete Gaussian curvature. And then, from these characteristic points the normal vector direction of the plane is calculated. The face pose is recovered from the normal vector direction. Experiment results demonstrate that the proposed approach has a low computation burden and wide applicability, and offers a good recovery effect.