Comparison of facial recognition methods based on extension methods of Principal Component Analysis

来源 :The 24th International Workshop on Matrices and Statistics(第 | 被引量 : 0次 | 上传用户:promethean65
下载到本地 , 更方便阅读
声明 : 本文档内容版权归属内容提供方 , 如果您对本文有版权争议 , 可与客服联系进行内容授权或下架
论文部分内容阅读
  The human face recognition system becomes more and more universal in different parts of our society like governments, banks and social welfare. How to improve the efficiency of the discrimination of human face is the most significant issue in the human face recognition system. And it is also a hot issue in the high dimensional analysis.In this paper, we compared the performance of the facial recognition of three methods, EMPCA, Sparse PCA and Kernel PCA. These three methods based on the general idea of Principal Component Analysis and Fishers Linear Discriminant. Our study based on the ORL face data base. EMPCA used same idea of the classic PCA by using the EM algorithm. The Sparse PCA extends the classic PCA by adding sparsity constraint in order to explain the variables more clear. The Kernel PCA extends the classic PCA by using kernel methods. The comparison of these three methods in facial recognition can help us know much more about the applications of the dimension reduction methods.
其他文献
  Methods of survival analysis (e.g., the Cox proportional hazards model) require that the event time be measured with respect to some origin time. The choice
会议
  In this paper we investigate methods for adjustment for covariate measurement errors in complex surveys. Focus is on the adjustment for covariate measuremen
会议
  In this talk block circular matrices which can be used to model dependence structure of hierarchical or multilevel data are considered. Several spectral pro
会议
  This paper investigates the M/M/s queuing model to predict an estimate for the proximity of the performance measures of queues with arrival processes that a
  The problem of testing the separability of a covariance matrix against an unstructured variance covariance matrix in the context of multivariate repeated me
会议
  For GI/G/1 queue, we transfer the problem of the law of iterated logarithm for queue length process, busytime process and the workload process into problems
会议
  The problems we consider in this talk arose in a probabilistic treatment of card shuffling. However we treat them as stochastic discrete time switching syst
会议
  Fitting quadratic curves (circle, ellipse, hyperbola and parabola) to given data points in the plane is a fundamental task in many fields like engineering,
  This paper focuses on trajectory analysis that applies finite mixture modeling to longitudinal data. The paper introduces new modeling tools using semiparam
会议
  In this work we consider the general linear model and some of its extensions. We study their sensitivity matrix results, with theore-tical developments and
会议