一种用于人脸识别的三维人脸模型重建方法

针对以往三维人脸模型重建算法实用性差、算法复杂度高、需要通用人脸模型和对噪声敏感等缺陷，提出了一种计算量小、无需通用人脸模型的三维人脸模型的重建算法。该算法在人工辅助确定特征点的基础上，利用能量函数最小的约束关系实现深度图的初步融合，然后运用改进ICP算法获得隐式的三维人脸模型。通过对获得模型的变换和投影，可产生不同姿态的二维人脸图像。实验结果表明，融合平均误差仅为1．32毫米，效果逼真。和其它算法相比，它还具有存储资源消耗少、算法稳定性高等优点。

A 3-D Face model reconstruction method for face recognition

Aimed at overcoming impracticability, high computation costs, requirement for general face model and sensitivity to noise in 3-D face model reconstruction, an algorithm using three range images is presented. This algorithm makes use of three feature points in the face and minimizing energy function to merge the range images coarsely. Then it applies the improved ICP algorithm to get the 3-D face model. The method reduces the space for storage. The computation cost of the algorithm is quite low. It does not need any general face model and is insensitive to the noise. Experimental results show that the algorithm could produce realistic 2-D face images. The average error of integration is about 1.32mm.