The comparison of heterogeneous samples extensively exists in manyapplications, especially in the task of image classification. In this paper, wepropose a simple but effective coupled neural network, called Deeply CoupledAutoencoder Networks (DCAN), which seeks to build two deep neural networks,coupled with each other in every corresponding layers. In DCAN, each deepstructure is developed via stacking multiple discriminative coupledauto-encoders, a denoising auto-encoder trained with maximum margin criterionconsisting of intra-class compactness and inter-class penalty. This singlelayer component makes our model simultaneously preserve the local consistencyand enhance its discriminative capability. With increasing number of layers,the coupled networks can gradually narrow the gap between the two views.Extensive experiments on cross-view image classification tasks demonstrate thesuperiority of our method over state-of-the-art methods.
展开▼
