Projection Analysis: Theory And Application In Face Recognition

Face recognition technology is a very active subject in the area of computer pattern recognition, which has a wide range of commercial and law enforcement potential applications. The primary task at hand, given still or video images, requires the identification of one or more persons using a database of stored face images.Based on Fisher's discriminant function, optimal sets of discriminant vectors have great influence in the area of pattern recognition. This paper first develops a theoretical framework for the Fisher linear discriminant analysis and shows it to be an improvement of the classical linear discriminant analysis in theory. The inherent relationship between Fisher linear discriminant analysis and Karhunen-Loeve expansion is revealed. Based on the analysis, this paper finally develops an optimal K-L expansion method. Moreover, this paper also extends it to its kernelized version. The proposed methods are tested on the ORL database, and the experimental results indicate that they are more effective than the other methods.The research on combination methods of classifiers is a hot spot in the area of computer pattern recognition. It has a wide range of applications, such as handwritten digit recognition, face recognition. Recently many combination rules have been used in face recognition. Based on Fisher discriminant criterion, In order to extract features by using the uncorrelated discriminant transformation, this paper used orthogonal wavelet transformation and K-L transformation to process the face images at first. According to people's recognition experience, we used multi-feature and multi-classifier combination to give out the classification results. Experiments on ORL database have obtained good results.In this paper, several different features are extracted from face image and fused. The complex vector is utilized to represent the parallel combined features and the linear projection analysis methods are generalized for feature extraction in the complex feature space. The experiments on NUST603 face image database indicate that the classification accuracy is increased after parallel feature fusion.In this paper, by developing a method for updating eigen decomposition, we proposed an incremental Principal Component Analysis. Experiment results on NUST603 face image databases show that this form of updating can still achieve comparable performance as the batch version.