具有网络一致结构的三维人脸模型重建研究

Research on 3D Face Model Reconstruction with Network Consistent Structure

针对传统方法及机器学习方法对大量三维人脸数据、训练样本数量与质量依赖性大的问题,采用基于光照立体的方法,利用人脸图像重建三维人脸。综合利用基于法线的模型变形法和非刚性变形法,提出一种基于法线的非刚性变形算法。利用SFS算法计算顶点法线,然后使用法线和局部刚性约束使参考模型变形,进而得到与参考模型具有一致网络结构的三维人脸模型,从而得到三维人脸图像。该方法与传统方法相比,节省了大量样本要求,并且立体效果更好。

Aiming at the problem that traditional methods and machine learning methods depend on a large number of 3D face data,the number and quality of training samples,this paper uses the method based on lighting stereo to realize the reconstruction of a threedimensional face from a face image.A non-rigid model deformation algorithm based on normal is proposed,which combines a normal based deformation algorithm and a non-rigid model deformation algorithm.The SFS is used to estimate the normal of the vertices,then updated normal are used to deform the reference face model with local rigid constrain,finally a deformed 3D face model is obtained,which has the same mesh structure with the reference model,thereby obtaining a 3D face image.Compared with the traditional method,the proposed method can save a lot of sample requirementsand the stereo effect is better.