A novel glasses-faces recognition method based on the 2D unsupervised geodesic discriminant projection (2DUGDP) technique is presented in this paper. Based on the virtual samples, discriminable features will be obtained by analyzing the difference of faces with variety eyeglasses. This feature characterizes the local scatter as well as the nonlocal scatter, seeking to find a projection that simultaneously maximizes the nonlocal scatter and minimizes the local scatter. The projection ensures the distance of samples remain close for near samples, and separate for far samples. Face space is regarded as a nonlinear instructure embedded in the high dimensional space. Geodesic distance is employed to model the intrinsic structure of the manifold. The method is applied to glasses-faces recognition and examination using the CAS-PEAL, FERET face databases. Results show that 2DUDP outperforms other methods.%针对戴眼镜人脸识别问题,提出了二维非监督测地线判别投影(2D unsupervised geodesic discriminant projection,2DUGDP)方法.该方法在扩充虚拟样本库的基础上,分析戴眼镜人脸图像和不戴眼镜人脸图像的差异及戴不同眼镜人睑图像的差异,提取判别特征用于识别.该特征考虑局部特征的同时考虑非局部特征,寻找一种最优投影在最大化非局部散度矩阵的同时最小化局部散度矩阵,使得距离近的数据投影后仍然近,距离远的数据投影后仍然远.考虑到人脸是非线性的流形结构,文中采用测地线距离表示样本间的差异.在FERET人脸库和CAS-PEAL人脸库上分别进行了实验,实验结果表明,该方法相比较其他方法更能提高戴眼镜人脸的识别率.
展开▼
