Message Coding and Compression with Artificial Neural Networks

Abstract

The need to overcome data preprocessing inherent in much of the classical data coding techniques commonly available led to the search for a free, easy-to-use, but flexible and powerful method. Artificial Neural networks have been attracting more and more researchers since the past decades. The distinct properties, such as learning ability,nonlinearity, fault tolerance, generalization etc., make it suitable for information protection, such as data encryption, data authentication, data detection, etc. In this paper a simple and low-cost coding method based on neural networks is proposed to be used to patterns compression. The goal of the developers is to build a tool able to store and send a coded and compressed message. The formed two-dimensional patterns are coded and compressed using the multilayer neural network with Back-propagation training algorithm. Hidden layer outputs of a trained network are sent as two-dimensional data,which represents the encoded vectors. To reconstruct the original patterns, this requires the output weights matrix and the output nodes functions which are unknown and not available in the encoded sent vectors. A compression rate of about 6:1 has been achieved.