Sparse Wavelet Auto-Encoders for Image Classification

The goal of the Deep learning methods is learning feature hierarchies with features from higher levels to lower level features of the hierarchy. The major contribution of this paper is to show how to extract features and train an image classification system on large-scale datasets. This method is an improvement of our recent work. The training is carried out by the combination of the most used methods for image classification: Deep Learning and the Wavelet Network. Some algorithms of DL like the sparse coding and the stacked autoencoders are used in our approach. For the WN, the Fast Wavelet Transform and the Best Contribution Algorithm are utilized. The ImageNet dataset used in the test phase in which we used many criteria such as the number of the hidden layers and the number of images that we specified for the training shows great efficiency of our model for image classification compared to another approach.