A Novel Steganography Approach Using S-CycleGAN With an Improvement of Loss Function

1. Introduction

The wide use of personal computers and the availability of multimedia content on the web provided the ideal conditions for the exposing of private information since information might contain valuable, personal, medical information (Hore et al. 2016; Chakraborty et al. 2017) or even confidential information (Roy et al. 2019; Seal et al. 2017). Furthermore, there are certain major risks to the leaked information, including unauthorized manipulation, copying, and transmission. Image steganography has evolved as a crucial matter in the prospect of secret communication and copyright infringement. Since steganography involves complex computations to embed secret information in cover images, implementing it for technology speeds up the process and increases its acceptability (Shet et al. 2019; Roy et al. 2020; Chakraborty & Mali, 2021a, 2021b; Singh et al. 2020). However, the cover image will still show the changes brought about by the embedding, allowing it to apply processes more effectively called steganalysis, a technique for finding secret data. The exchange of information between the mobile unit and the server creates the interface between the items and the network, which sends information to the mobile terminal over the network (Qiu et al. 2018; Chakraborty et al. 2016; Mali et al. 2021). Public service computing in this context is typically provided by high-performance servers (Qiu et al. 2018). The emergency packets are now being used and upgraded to effectively handle the network congestion issue in the Internet of Things, we propose a novel steganography method based on generative adversarial networks (GAN). If the cover data are available to the public, a third party can easily decode the secret message by comparing the cover data to the message being hidden. We require a distinct cover data for each piece of secret communication in order to prevent this problem. In this study, we rely on images as the cover data. We can employ GAN to create a great deal of unique and authentic-looking photos. In reality, GAN training leads to the creation of the generator and discriminator neural networks. The discriminator evaluates the virtual pictures' naturalness once the generator generates them. The generator and discriminator are both used in the proposed methods to secure the sender side cover information is authentic and to remove stego data provided by unreliable third parties. In this work, before considering the viability of using the discriminator we need to test naturalness of images, we first established that the generator can produce an infinite amount of cover data. We hope that the proposed method can provide a more effective way of information concealment.