中文摘要：

提出一种基于核方法的无监督鉴别投影,在较好地描述人脸图像的同时,对图像进行有效地分类。对核局部保留投影（KLPP）和无监督鉴别投影技术（UDP）进行了相应的研究,将两者互相结合。该方法同时考虑到样本的局部特性和非局部特性,有效地利用了对分类有用的重要信息;此外,将核方法和流形学习方法结合起来,有效地描述人脸图像的非线性变化,对于人脸识别问题有较好的效果。在Yale库上的实验表明,该方法的识别率明显高于UDP和PCA,且有较好的分类效果。

英文摘要：

A method called kemel-based unsupervised discriminant projection is proposed. It not only describe the human face effectively, but also have a good effect in classification. To the corresponding research on kernel locality preserving projection （KLPP） and unsupervised discriminant projection （UDP）, they are combined with each other. Both the local characteristics and nonlocal characteristics of samples are taken account, which uses the effective information that is important to classification. Besides, it combines the kernel trick with the manifold learning method, and it can describe the nonlinear change of the face effectively, has a good effect on face recognition problem. The experiment on Yale face database shows that the proposed method outperforms UDP and PCA. Besides, it has a good effect in classification.