Font Size: a A A

A Weighted Kernel PCA And The Related Parameters Choice

Posted on:2010-10-16Degree:MasterType:Thesis
Country:ChinaCandidate:W Q MaFull Text:PDF
GTID:2178360275953640Subject:Applied Mathematics
Abstract/Summary:PDF Full Text Request
As the development of the society, there are increasing demands in automatic identity check. Since some biological characteristics are intrinsic and stable to people and strongly different from one to the others, they can be used as features for identity check. Among all the characteristics of human, the characteristics of face are the most direct tools which are friendly and convenient and can easily be accepted by the customers.Face recognition is an extensive and challenging research topic. Recently, significant progresses have been made in the technology of the face recognition. And feature extraction is the critical stages in the face recognition. Principal Component Analysis is acknowledged one of the most powerful techniques for feature extraction. As a nonlinear form of Principal Component Analysis, Kernel Principal Component Analysis has been broad applied to face recognition in recent years. However, each dimension feature of face images is treated equally by the principal component analysis and kernel principal component analysis in feature extraction. But in fact, different features play different roles in face recognition. For example, the eyes and mouth play far greater importance role than cheeks, the former part of face is greater importance than the lower part of face, with special features of face such as mouth askew, cross-eye, and so on, to be identified more easily. In this paper, we adopt Gaussian distribution function as a weighted function, which can give prominence to the key features in face recognition. The proposed method is combined with the kernel principal component analysis to calculate the weighted subspace. Experimental results on the normal ORL face database show the proposed method is effective.Face recognition relates to these critical steps that are options of kernel function, parameter of kernel function, train sample and classifier. Many researchers probe to these problems, but no better methods can instruct how choose optimization parameter. At present, many methods confine given applied field via experiment specifies relevant parameter. So this article receives the best range of relative parameters via changing parameters of weighted kernel function and kernel function. And obtaining the best rate of identification is its object.
Keywords/Search Tags:Weighted Kernel Principal Component Analysis, Kernel Principal Component Analysis (KPCA), Face Recognition, Kernel Function, Cosine Distance Classifier
PDF Full Text Request
Related items