Fig. 2From: A semi-supervised convolutional neural network based on subspace representation for image classificationConceptual illustration of the proposed shallow network. DFSNet employs two distinct filters banks which work in complementary directions. In order to reduce the high dimensionality of the features and increase rotation invariance, the proposed method is followed by a feature mapping, as is done in most of the shallow networks. Then, the classification is performed by linear support vector machineBack to article page