Face recognition has aroused great concern for decades since it serves as a significant part in the fields of human-machine interaction as well as bioinformatics. Facial feature extraction is one of the key steps in face recognition system However

this step is characterized as being easily influenced by variations in face images such as illumination condition and expressions. In order to address this problem

a method that utilizes（2D)2PCA to extract facial features on the sub-bands obtained via wavelet packet decomposition（WPD) is proposed. There are three contributions:（1) take all multi-resolution sub-bands as research objects;（2) choose ‘successful’ sub-bands based on recognition rates;（3) propose a sub-band fusion method. Firstly

sub-bands are acquired by two-level WPD

then the feature matrixes of all sub-bands are calculated by（2D)2PCA

and further used to obtain recognition rates with the nearest neighborhood classifier. Thirdly

‘successful’ sub-bands are chosen based on their recognition rates and fused to complete the task of face recognition. Finally

intra and extra experimental comparisons using samples of CMU PIE and Yale indicate that the proposed method gain satisfactory results and fusing sub-bands on multi-resolution can improve recognition performance.