基于二维有监督测地线判别投影的戴眼镜人脸识别

2D Supervised Geodesic Discriminant Projection based glasses-face recognition

针对戴眼镜人脸识别问题,提出了二维有监督测地线判别投影(2D Supervised Geodes-ic Discriminant Projection,2DSGDP)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,及戴不同眼镜人脸图像的差异,提取判别特征用于识别。该特征同时考虑类内类间差异,寻找一种在最大化类间散度矩阵的同时最小化类内散度投影矩阵,使得属于同一类的数据投影后距离近,不同类的数据投影后距离远,降维投影后的类别特征得以保持。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。

A novel glasses-face recognition method based 2D Supervised Geodesic Discriminant Projection（2DSGDP） technique is presented.Based on the virtual samples,it is analysied the difference of between glasses-face and face,the difference of between different glasses,then get discriminate feature to recognition.These features characterizes the within scatter as well as the between scatter,seeking to find a projection that simultaneously maximizes the between scatter and minimizes the within scatter,so that the data belong to the same class become nearly,belong to the different class become far.The proposed methods are applied to glasses-faces recognition.The experimental results show that 2DSGDP outperforms other methods in the CAS-PEAL and FERET face database.