Face Time - Varying Nonlinear Classification Algorithm Of Feature Extraction And Nuclear

Face recognition is to extract personal features from static image(s) or video sequence containing face(s), to confirm or identify the identification of the object. A completed face recognition system usually involves of multiple functional modules, such as face detection or tracking, image preprocessing, feature extraction and selection, and classification. In realistic and complex environments, the key and the most challenge step for successful face recognition is to extract reliable and robust personal features.This thesis mainly focuses on face feature extraction and classification algorithms.As for feature extraction, the most commonly used algorithms, Principal Components Analysis (PCA) and Linear Discriminant Analysis (LDA), are discussed and compared, and then the concept of time-frequency analysis is introduced to feature extraction and some commonly used time-frequency distributions (TFDs) are discussed. As a special TFD, the definition and traits of Fractional Fourier Transform (FrFT) is then introduced, and a definition and calculation algorithm for discrete FrFT (DFrFT) is listed for face feature extraction application in the sequel.In classification, some of the commonly used classifiers are categorized into linear and nonlinear groups, based on which a classifier called Kernel-based Nonlinear Representor (KNR) is introduced for optimal feature representation.Finally, experiments are taken to illustrate the feasibility of DFrFT-based feature extraction and KNR-based classification, using the famous ORL face database.