Deep learning with shallow architecture for image classification

Abstract

This paper presents a new scheme for image classification. The proposed scheme depicts a shallow architecture of Convolutional Neural Network (CNN) providing deep learning: For each image, we calculated the connection weights between the input layer and the hidden layer based on MultiResolution Analysis (MRA) at different levels of abstraction. Then, we selected the best features, representing well each class of images, with their corresponding weights using Adaboost algorithm. These weights are used as the connection weights between the hidden layer and the output layer, and will be used in the test phase to classify a given query image. The proposed approach was tested on different datasets and the obtained results prove the efficiency and the speed of the proposed approach.