Image compression with artificial neural networks

Abstract

In this paper, we make an experimental study of some techniques of image compression based on artificial neural networks, particularly algorithm based on back-propagation gradient error [5]. We also present a new hybrid method based on the use of a multilayer perceptron which combines hierarchical and adaptative schemes. The idea is to compute in a parallel way, the back propagation algorithm on an adaptative neural network that uses sub-neural networks with a hierarchical structure to classify the image blocks in entry according to their activity. The results come from the Yann Le Cun database [7], and show that the proposed hybrid method gives good results in some cases. © 2013 Springer-Verlag Berlin Heidelberg.