基于支持向量回归机的三维人脸建模方法研究

提出一种新的三维人脸建模方法,该方法首先运用双目立体视觉原理从不同角度获得人脸的两张照片,在一张照片上选定特征点,通过匹配在另一张照片上得到对应的特征点,然后从空间立体几何知识出发,根据两张照片上特征点的坐标以及拍摄时的参数,计算其三维坐标,利用支持向量回归机(SVR)对其进行回归预测建模,最后经纹理贴图,得到具有纹理特征的特定三维人脸模型。仿真实验结果表明,该方法获得的三维人脸模型较为逼真。

3D Face Modeling Method Based on SVR

This paper presents a new 3D face modeling method. First, it needs to shoot two face photos of different angles applies the principle of binocular stereovision. Then we select some feature points in one photo and obtain the corresponding feature points in another by matching. And we calculate the 3D coordinates according to the coordinates of the feature points in the two photos as well as the shooting parameters using the three-dimensional apace geometry knowledge. Followed, we modeling the 3D face with the method of fitting using Support Vector Regression Machine (SVR).At the end, the specific 3D face model with texture features is generated after the texture mapping. Simulation results show that the three-dimensional face model obtained by this method is more realistic.