Steganography GAN: Cracking Steganography with Cycle Generative Adversarial Networks

TL;DR

This paper demonstrates that CycleGANs combined with Bayesian Optimization can effectively crack LSB steganography, outperforming convolutional autoencoders and opening new research directions.

Contribution

The study introduces a novel approach using CycleGANs to crack steganography, showing its effectiveness over traditional autoencoder methods.

Findings

CycleGANs successfully crack LSB steganography.

CycleGANs outperform convolutional autoencoders in this task.

The approach opens new avenues for steganography detection and analysis.

Abstract

For as long as humans have participated in the act of communication, concealing information in those communicative mediums has manifested into an art of its own. Crytographic messages, through written language or images, are a means of concealment, usually reserved for highly sensitive or compromising information. Specifically, the field of Cryptography is the construction and analysis of protocols that prevent third parties from understanding private messages. Steganography is related to Cryptography in that the goal is to obscure information using some method or algorithm, but the most important difference is that the information and the method of concealing information within Steganography both involve images--more precisely, the embedding of one image or piece of information into another image. Ever since the creation of covert communication methods, steps have been taken to crack cryptography and steganography algorithms. The desire for this rises from both human curiosity and the need to counteract adverse uses, such as encoding harmful media in inconspicuous media (phishing attack). In this paper, we succeed in cracking the Least Significant Bit (LSB) steganography algorithm using Cycle Generative Adversarial Networks (CycleGANs) and Bayesian Optimization and compare the use of CycleGANs against Convolutional Autoencoders. The results of our experiments highlight the promising nature of CycleGANs in cracking steganography and open several possible avenues of research.