Linear And Non-linear Feature Extraction Measures At The Application Of Face Recognition

From the view of extracting face features, we discussed two sorts of face recognition methods in this paper,such as linear subspace method and non-linear subspace methods. Among the non-linear subspace methods, we focus our attention on the principles and algorithms belonging to manifold learning methods. After making a deep research, we develop some new algorithms as regards it and these algorithms are verified to be effective in the application of face recognition.Firstly,we discuss linear subspace method in the paper.We firstly explore the unsupervised linear feature extraction methods,such as Principal Component Analysis(PCA) and Independent Component Analysis(ICA),In the next we analyze the supervised linear feature extraction methods,such as Linear Discriminant Analysia(LDA)and Maximum Margin Criterion(MMC),in the article.After having a deep study on their principles and algorithm, we try these methods out on the ORL facebase and Yale facebase.The results show supervised linear subspace methods are more effective than unsupervised linear subspace methods.Secondly,non-linear face feature extraction methods are researched. In the article we mainly focus on manifold feature extraction methods.We interpret their principles and algorithm of these methods.The classical manifold learning methods have isometric mapping(ISOMAP),locally linear embedding(LLE),laplacian eigenmap(LE),locality preserving projections(LPP) and unsupervised discriminant projection(UDP). Through using different classifiter experiments on the different facebases,we discover their characteristics on the application of face recognition.At last,we proposes modified UDP algorithms called Orthogonalized UDP and Uncorrelative UDP. We prove feature equation of these modified UDP algorithms and apply them to face recognition. Through doing experiment on the different facebases, we verify that they can reach better recognition performance comparing with UDP.