Effectiveness of various classification techniques on human face recognition

In this paper the effectiveness of different classification techniques is evaluated on the performance of face recognition algorithms. Gabor wavelet and its fusion with local binary pattern (LBP) are utilized as feature extractors. Dimensionality reduction approaches, principal component analysis (PCA) and Fisher's linear discriminant (FLD), are employed to reduce the size of feature vector. The performance of nearest neighbor (NN) classifier with various cost functions, sparse classification, multilayer feed-forward neural network (MFNN) and extreme learning machine (ELM) are analysed on three face databases, Extended YaleB, FERET and Multi-PIE, which contain large number of individuals with images under various illumination conditions and different facial expressions. Simulation results show that ELM and MFNN are effective in all conditions. The performance of nearest neighbor and sparse classifier is degraded under severe illumination variation.