基于拟牛顿和遗传算法的三维人脸建模

3D face modeling based on quasi-Newton and genetic algorithm

用网格重采样算法和点变形技术把不同人脸纹理全景图进行标准化,实现三维人脸的像素级对齐;根据在不同姿态下用拟牛顿法优化目标函数优化速度的不同和提取的人脸特征点精确计算出输入图像中人脸的姿态;确定三维人脸模型在此姿态下的可见点,最后利用改进的实数遗传算法进行匹配计算,建立完整的三维人脸模型。实验表明,此算法能实现三维人脸像素级的精确对齐,快速的精确计算输入图像中人脸的姿态,减少优化参数,简化目标函数,提高模型匹配效率和重构精度,缩短匹配时间。

Pixel-pixel correspondence of different face textures are achieved by transform all face panoramic texture images into standard template using mesh resampling and points warping. According to the difference of objective function optimization speed by Quasi-Newton method under different pose, the face pose of the input image is precisely calculated with feature points. The 3D face model is reconstructed by improved real-code genetic algothrn after determined the visible points. Experimental results show that this algorithm can realize the precise pixel-pixel correspondence of 3D face models, quickly and accurately calculate face vase, reduced optimization parameters, simplify the objective function, improve the efficiency and precision of model matching, and shorten the match time.