基于样例学习的任意光照下的人脸3D重建方法

Example-based learning method for 3D face reconstruction under varying lighting

待匹配的人脸图像与数据库中的原型图像之间的光照差异是自动人脸识别的主要瓶颈问题之一。提出了一种基于样例学习方式的3D人脸形状重建方法,既可以生成任意光照条件下的数据库中人脸图像,也可以对待识别图像进行重新光照,合成无阴影的图像。该方法在建立人脸数据库时利用光度立体技术分离人脸图像的纹理和形状信息,并用多面体模型在最小二乘意义下恢复其3D信息并更新法向量场以克服阴影误差,从而可以利用计算机图形学的方法合成任意光照条件下和小角度姿态改变时的人脸图像;在识别时采用数据库中3D数据的线性组合形式对输入图像建模,以估计其3D信息,从而可以重新照明。在YaleB人脸数据库上的实验表明,在建立3D人脸数据库后,该方法可以快速恢复输入单幅图像中人脸的3D信息,并生成任意光照条件的该人脸图像。

Illumination changing from picture to picture is main barrier toward full automatic face recognition.In this paper,a novel method to handle lighting conditions is proposed,which can construct an image of human face in database under arbitrary illumination conditions and also can relight the probe image to no shadow image of human face.We carry out the method by decomposing the＂texture image＂ and the＂3D shape＂ based on the photometric stereo technology,and then using the model of polyhedron to get 3D shape of human face by least-squares-fitting.The 3D shape can be used to update the texture to conquer the influence of cast shadow.When the probe image is inputted,the linear combine of normal vector field and texture in example-database can be used to fit the probe image and then relight the image or reconstruct the 3D shape of human face in probe image.Experimental results on YaleB database show that this method can reconstruct the 3D shape of human face surface in probe image effectively,and synthesize the probe human face image under arbitrary illumination conditions.