关键词: 核方法, 多元t分布, 多元正态分布, 贝叶斯分类, 人脸识别

Abstract: In order to solve the problem of multivariate normal distribution failing to comply with the distribution of sample data when it has a serious tailing phenomenon, a multiclass classification method under t distribution was proposed. Sample data were extended to the high dimensional feature space by kernel method and Maximum A Posteriori (MAP) was obtained by Bayesian classifier, and then classification result could be gotten. Because it has another degree of freedom parameter v in multivariate t distribution, it can more easily capture the complexity of the sample data, and enjoys better robustness. A large number of experiments have been done on the five international standard UCI data sets and three facial image data sets, and the experimental results show that the method achieves better classification and is feasible.

Key words: kernel method, multivariate t distribution, multivariate normal distribution, Bayesian classification, face recognition