Integrating Companding and Deep Learning on Bandwidth-Limited Image Transmission

Abstract

The image companding is a simple image compression technique which is very easy to be implemented in the bandwidth-limited environment. This paper presents a simple way for improving the quality of decompressed image in the image companding task. The proposed method consists of two networks, namely Sub-band Network (SubNet) and Pixel Network (PixNet), for performing an image reconstruction. The SubNet module exploits the effectiveness of Stationary Wavelet Transform (SWT) and Convolutional Neural Network (CNN) in order to recover the lost information in the wavelet sub-bands basis. Whilst, the PixNet part applies CNN with identity mapping to improve the quality of initial reconstructed image obtained from the SubNet module. As reported in this paper, the proposed method outperforms the former existing schemes in the image companding task. It has also been proven that the proposed method is able to improve the quality of reconstructed image with some simple steps.