Improvement in PCA Performance Using FLD and RBF Neural Networks for Face Recognition

Face is an important biometric feature for personal identification. This paper describes a new face verification method based on singular value decomposition and RBF neural networks. The proposed method utilizes the positive samples and negative samples learning ability of RBF neural networks to improve the principal component analysis (PCA) based face verification. Experiment results show that the novel face verification method is effective and possesses several desirable properties when it compared with many existing methods. The face features are first extracted by PCA method. The Fisher Linear Discriminant (FLD) is commonly used in pattern recognition. It finds a linear subspace that maximally separates class patterns. The resulting features from PCA are further processed by the Fisher´s linear discriminant (FLD) technique to acquire lower-dimensional discriminant patterns. Radial basis function (RBF) neural classifier is used to cope with small training sets of high dimension, which is a problem frequently encountered in face recognition. Simulation results conducted on the YALE face database show that the system achieves excellent performance in terms of recognition accuracy of 92%.