Research On Feature Extraction Algorithms In Face Recognition

Face recognition is one of the most important branches of biometrics, and it is widely applied in certificate validation, access control systems, video surveillance system and criminal detection, etc. Feature extraction plays an important role in the area of face recognition. The classification performance of a face recognition system greatly depends on the extracted features, which should contain sufficient discriminant information residing in the original data. Combining the existing feature extraction theory with optimization algorithms or transformation methods is a primary research field for effective feature extraction. Feature extraction theory with manifold methods is further discussed in this thesis, in which some proposed algorithms work well for face recognition.The primary contributions of this thesis are listed as follows:(1) A new feature extraction method based on chaos genetic algorithm (CGA) and principal component analysis (PCA) is proposed. The chaos genetic algorithm uses two kinds of chaotic mappings in different ways, which maintains the diversity of population and enhances the global searching ability. Then CGA is used in feature (eigenvector) selection after the transformation of PCA, which can quickly find out feature subspace that is most beneficial to classification. The experimental results on ORL face database indicate that the proposed method not only reduces the dimensions of face feature space, but also achieves higher recognition performance than other methods.(2) A feature extraction method based on discrete cosine transform (DCT) and fuzzy linear discriminant analysis (FLDA) is given. The face images are denoised by DCT, and dimension reduced features are obtained, then the FLDA is performed on the feature vectors to enhance dicriminant power. Finally, the minimum distance classifier is selected to perform face classification. The experimental results on ORL face database demonstrate that the proposed method can effectively filter the high frequency interference except face image information, strengthen the characteristics of the dicriminant power and obtain an ideal recognition results.