In pattern recognition

combining multiple discrepant and complementary classifiers can improve the accuracy and stability of identification system. Bagging

boosting and random subspace methods are commonly used to combine multiple weak learners. A random sampling subspaces locality preserving projections (RSSLPP) method is proposed to improve the recognition performance of a single locality preserving projections (LPP) in this paper. At the training stage

several random sampling principle components subspaces are generated by random sampling the principle components subspace of training set. Then

multiple discrepant and complementary locality preserving projections subspaces are generated by applying locality preserving projections method to the projected samples of training set on the random sampling principle components subspaces. At the recognition stage

test sample is successively projected into each random sampling principle components subspaces and the corresponding LPP subspaces. On each LPP subspaces

the nearest neighbor classifier is used for classification. Finally

majority voting criterion is used to fuse the recognition results of each LPP subspaces. Experiments on FERET subset illustrate that random sampling subspaces locality preserving projections method effectively combines the complementary information of each LPP subspaces and improves face recognition accuracy.