Aiming at recognition problem caused by variations in lighting

facial expression etc and small sample size problem which generally exist in face recognition

this paper proposes a novel recognition method

called RB2DLDA

which fuses bidirectional 2DLDA feature by resampling technique.In matrix-based LDA

2DLDA uses inner-class and intra-class covariance information in the vertical direction

and E2DLDA uses inner-class and intra-class covariance information in the horizontal direction

this paper proves in theory that class information in the two directions is complementary to each other to some extent

provides theoretical foundation for fusing class information in the two directions to design classifier and improves recognition performance of the classifier.Moreover

proposes an adaptive algorithm of setting feature reduction parameter.Our method is tested on AR and CAS-PEAL-R1 face databases

and the experimental results show that the proposed approach possesses higher recognition accuracy and robustness than other state-of-the art methods.