单张正面人脸图像的三维人脸重建方法研究

3D Face Reconstruction Based on Single Frontal Face Image

针对基于单张正面人脸图像进行三维人脸重建时所需脸部侧面深度信息缺失的问题,提出基于BP神经网络快速三维重建方法。通过建立BP神经网络估计出正侧面人脸数据的关系,从而由输入的正面数据得到侧面数据,并对BP算法做出改进,加速了算法的收敛,提高了拟合的精度。然后利用获取的人脸侧面数据调整CANDIDE-3人脸模型,生成近似图像中人脸的目标几何模型,最后通过纹理映射生成具有真实感的特定三维人脸模型。实验证明,这种重建方法快速简便,真实感强,较好地解决了基于单张照片重建时侧面深度信息缺失的问题。

According to the problem of missing side-face depth information on 3 D face reconstruction from a single frontal-face image, a quick 3D reconstruction approach is proposed based on BP Neural Network. The approximate relationship between frontal-face data and side-face data can be estimated through establishment of BP neural network. Then side-face data can be obtained from input of frontal-face data. An improved BP algorithm is proposed. Then the convergence speed and the fitted accuracy are both increased. A tbarget geometric model approximale to the face in the image is constructed by modifying CANDIDE-3 model tased on frontal-face data and side-face data obtained before. Finally an individual realistic 3D face model is generated through texture mapping. The experiments show that this approach can solve the problem of missing side-face depth information well and it is fast and efficient, the result is realistic.