基于最大边距局部敏感鉴别分析的人脸识别

Maximum Margin Neigborhood Preserving Embeding for Face Recognition

局部敏感鉴别分析(LSDA)是一种基于向量学习的提取特征的算法,该算法使得属于同一类的相邻数据经投影后尽量靠近,但不同类的邻近数据则相远离.在实际应用中,由于小样本问题,通常先利用PCA算法对原始数据进行降维处理,然后再使用LSDA算法提取特征.然而,这种方法会丢掉一些重要的鉴别信息.提出了最大边距局部敏感鉴别分析(MM-LSDA)算法,直接从原始数据中提取特征,避免了鉴别信息的损失,同时使得同类中的近邻数据尽量靠近,而不同类之间的样本远离.在ORL和Yale人脸库上的仿真实验表明此算法更有效.

Locality Sensitive Discriminant Analysis （LSDA） was a liner manifold learning algorithm, it makes that the nearby data points with the same label are closed to each other while the nearby points with dif- ferent labels are far apart. In practice, because of small sample size problem, PCA is applied to reduce the di- mension of original data space before utilizing LSDA. However, this strategy may discard important discrimina- tive information. Maximum Margin Locality Sensitive Discriminant Analysis （MM-LSDA） is proposed to over- come the above problem in this paper. The new method extracts features directly from original data, preserving the nearby data points with the same label, while aparting data points with different labels. Experiments on ORL and Yale databases show that the proposed method is more effective.