Abstract: Orthogonal neighborhood preserving projections (ONPP) is a typical graph-based dimensionality reduction technique,which preserves not only the locality but also the local and global geometry of the height dimensional data,and has been successfully applied to face recognition.The supervised ONPP tries to find the optimal embedding of low-dimensional subspace by setting up homogeneous adjacency graphic and minimizing the homogeneous local reconstruction errors.However,it only uses the homogeneous information,which leads to unconspicuous structure of heteroge-neous data.Motivated by this fact,we proposed a novel method called double adjacency graphs based orthogonal neighborhood preserving projections (DAG-ONPP).By introducing homogeneous and heterogeneous neighbor adjacency graphs,the homogeneous reconstructing errors will be as small as possible and the heterogeneous reconstructing errors will be more obvious after data being embedded in low-dimensional subspace.The results of the experiments on the ORL,Yale,YaleB and PIE databases demonstrate that the proposed method can markedly improve the classification ability of the original method and outperforms the other typical methods.

Key words: Supervised learning,Face recognition,Manifold learning,Orthogonal neighborhood preserving projections,Double adjacency graphs