基于三维数据与MMSV特征的二维人脸识别*

针对二维人脸识别对姿态与光照变化较为敏感的问题,提出了一种基于三维数据与混合多尺度奇异值特征MMSV(mixture of multi-scale singular value,MMSV)的二维人脸识别方法。在训练阶段,利用三维人脸数据与光照模型获取大量具有不同姿态和光照条件的二维虚拟图像,为构造完备的特征模板奠定基础;同时,通过子集划分有效地缓解了人脸特征提取过程中的非线性问题;最后对人脸图像进行MMSV特征提取,从而对人脸的全局与局部特征进行融合。在识别阶段,通过计算MMSV特征子空间距离完成分类识别。实验证明,提取到的MMSV特征包含有更多的鉴别信息,对姿态和光照变化具有理想的鲁棒性。该方法在WHU-3D数据库上取得了约98.4%的识别率。

2D face recognition based on 3D data and MMSV features

With regard to the problem of 2D face recognition is sensitive to pose and illumination variations, this paper proposed a novel approach, which was based on 3D data and MMSV features. In the training stage, used 3D face data and illumination model to generate a large number of 2D virtual images with varying pose and illumination in order to set up complete features template, and these virtual images were grouped into different subsets subsequently so as to relieve the nonlinear problems in feature extraction of human face. It extracted MMSV features at last to fuse global and local features. Recognition was accomplished by calculating the distance of MMSV feature subspace. The experiment results confirm that MMSV features contain more identifying information and the robust to pose and illumination variations. This approach achieved 98.4% recognition rate in WHU-3D database.