非线性流形上的性别识别算法研究

A Gender Recognition Algorithm Based on the Nonlinear Manifold

在高维非线性空间中,如何更有效地提取人脸图像的主要特征,以及如何更有效地区分不同的性别类别,已经成为性别识别中广泛关注的问题。针对这一问题,提出一种非线性流形上的性别识别算法。该算法不但能有效提取高维空间中数据点的主要特征,并且能充分挖掘出数据流形间的几何结构和判别结构,从而使不同性别之间达到最优化分类。通过ORL和Yale两个人脸数据集实验,并与PCA(Principal Components Analysis)+LDA(Linear Discriminant Analysis),PCA+SVM(Support Vector Machine),KPCA+LDA,KPCA+SVM 4种常用的性别识别算法进行比较。实验结果显示:所提出的算法与其他传统算法相比具有更高的识别率,且有一定的鲁棒性和较高的运行效率。

Abstract： The problem of effectively extracting the main feature of face image and distinguishing between different gender categories. In higher dimensional linear space has become a widespread concern in gender recognition community. This paper proposes a nonlinear manifold on gender recognition algorithm to solve this problem. It can not only effectively extract the main characteristics of the data points in high dimensional space, but also be fully excavated the geometric structure and the discriminant structure in the data manifold, and then achieve the gender classification optimization. Experiments were carried out through two face data sets of ORL and Yale and comparing with four kinds of combination algorithms like PCA（ Principal Components Analysis） ＋ LDA（ Linear Discriminant Analysis）, PCA（ Principal Components Analysis） ＋ SVM（ Support Vector Machine）, KPCA（ Kernel Principal Components Analysis） ＋ LDA（ Line- ar Discriminant Analysis）, KPCA（ Kernel Principal Components Analysis） ＋ SVM（ Support Vector Machine）. The experimental results show that the proposed algorithm presented a better recognition performance on gender recognition and a higher recognition rate than the other commonly algorithm. Meanwhile, it has achieved robustness with high operating efficiency. Key words： gender recognition; nonlinear manifold; geometry structure; diseriminant structure; nonlinear high-dimensional space