In view of face recognition with only on sample problem

we propose a weighted modular 2DPCA method in this paper. In the method

we first divide original images into modular images and accomplish the sub image 2DPCA feature extraction. Then

we use optical flow between testing and sample image to estimate difference of corresponding pixel blocks quantitatively

which is as criterion for us to give variant weights to each block of difference matrixes between the feature matrixes of sample and that of probing images. Finally

nearest neighbor classifier is employed for classification. The experiment results on the JAFFE and ORL human face database indicate that weighted modular 2DPCA is superior to both conventional 2DPCA and modular 2DPCA in terms of accuracy and robustness with the same dimension of discriminate features

and it is feasible to introduce prior knowledge into PCA method of face recognition.