Face Recognition Based on Deep Autoencoder Networks with Dropout

Though deep autoencoder networks show excellent ability in learning feature, its poor performance on test data go against visualization and classification of image. In particular, a standard neural net with multi-hidden layers typically fails to work when sample size is small. In order to improve the generalization ability and reduce over-fitting, we apply dropout to optimize the deep autoencoder networks. In this paper, we propose face recognition based on deep autoencoder networks with dropout. Our experiments show that deep autoencoder networks with dropout yield significantly lower test error, and bring a new conception in pattern recognition with deep learning.