In this paper, we propose a new deep learning method called Deep PCA (DPCA) for face recognition. Our method performs deep learning through hierarchically projecting face image vectors to different feature subspaces and obtaining the representations from different projections. Specifically, we perform a two-layer ZCA whitening plus PCA structure for learning hierarchical features. The whole feature representation of each face image can be extracted by concatenating the representations from the first and second layers. Our approach learns deep representations from the data, by utilizing information from the first layer to produce a new and different representation, making it more discriminative. Experimental results on the widely used FERET and AR databases are presented to show the efficiency of the proposed approach.
展开▼