Face recognition using 2DGabor mean values and local features fusion

A novelty method of face recognition combined with the fusion of local 2DGabor mean values and 2DPCA dimension deduction based on subspace analysis is proposed. Firstly, each facial image in the training sample set is divided according to the five special face regions and then the features of five key regions are extracted through 2DGabor wavelet, mean values are calculated from feature vectors gained from the corresponding pixel of every test sample and then the eigenvectors are obtained, secondly, 2DPCA is used to decrease the dimension of the gained eigenvectors, finally the nearest neighbor classification method is adopted to recognize the face images. The numerical experiments on face database of ORL, YALE and FERET show this method achieves better effect on face recognition than other methods and shows stronger robustness to changes of illumination, expressions, poses and so on.