Research On Kernel-based Nonlinear Feature Extraction And Image Recognition

Feature extraction is the elementary problem in the area of pattern recognition. For image recogniton tasks, extraction of effective image features is the key step. Kernel-based feature extraction methods are a very effective nonlinear feature extraction method which is recently proposed by vapnik. In this paper, we focus our attention on kernel-based nonlinear feature extraction and develop some new algorithms as regards it. And, these algorithms are verified to be effective in the application of face identification and handwritten characterrecognition.It is well known that Fisher-based discriminant analysis methods are the most effective feature extraction method.The kernel-based statistical uncorrelate discriminant analysis is proposed in the paper. The proposed method don't only extract nonlinear features effectively, but also removes statistical correlation between components of the pattern samples. On this basis, a fast algorithm for extraction of the optimal nonlinear discriminant vectors in high dimensional feature space is presented. Unlike the KFDA method, the proposed algorithm only need to perform in a low dimensional transformed space, which leads to significant computational reduction. The experimental results on ORL face database verify that the presented approach is faster than KFDA in terms of recognition time while retain their accuracy.Feature extraction is one of the most essential problems in pattern recognition. Kernel Fisher discriminant analysis (KFDA) is thoroughly studied in the paper. A equivalent but more simple nonlinear feature extraction method is found. The main idea of this method is that the original input space Rn is transformed into a lower dimensional feature space RN with N the number of the training samples and Nn, in which linear Fisher discriminant analysis is performed for extraction of optimal discriminant features. On this basis, a general model for feature extraction is proposed, by which, a matrix similarity based feature extraction algorithm is developed. Finally, the experimental results on ORL indicate that the proposed model is effective.A Two-stages kernel feature extraction methods for Face Recognition is developed in this paper. The algorithm includes two stages: firstly, the classical principal component analysis(C-PCA) is employed to condense the dimension of image vector. What follows, kernel Fisher disoriminant analysis(KFDA) or KPCA are applied to the reduced dimensional training samples. On this basis, a more efficient method, called I-PCA+KFDA/I-PCA+KPCA, are proposed. Different from the previous method where C-PCA is based on vectors, I-PCA is to exploits image matrices to directly construct theimage total scatter matrix. Finally, The experimental results on ORL face databases indicate that the proposed methods is more efficent than KFDA while retaining the same recognition accuracy.Clustering is a unsupervised learning technique widely applied to the field of pattern recognition. To solve the scalability for kernel methods, A novel method for dimensionality reduction of kernel matrix is presented in the paper. Its main idea is that the original input space is first transformed to a high dimensional feature space via a nonlinear mapping . The k-means clustering algorithm is used to reduce the number of the transformed training samples and a reduced set, also called representative set, is derived. Based on the representative set, a set of orthogonormal basis vectors is obtained and used to form a new lower dimensional projection subspace. The proposed method is tested on CENPARMI handwriting digit database of Concordia University. The experimental results show the proposed method can significantly reduce complexity of the found classifiers while retaining their accuracy.The paper developes a novel nonlinear feature extraction method based on wavelet features. Its main idea is that wavelet transform is first employed to preprocess the original training images before the nonlinear mapping and three groups of wavelet features, lowest frequency...