Exploiting deep neural networks for digital image compression

Deep neural networks (DNNs) are increasingly being researched and employed as a solution to various image and video processing tasks. In this paper we address the problem of digital image compression using DNNs. We use two different DNN architectures for image compression i.e. one employing the logistic sigmoid neurons and the other engaging the hyperbolic tangent neurons. Experiments show that the network employing the hyperbolic tangent neurons out performs the one with the sigmoid neurons. Results indicate that the hyperbolic tangent neurons not only improve the PSNR of the reconstructed images by a significant 2~5dB on average but they also converge several order of magnitude faster than the logistic sigmoid neurons.