Robust steganographic approach using generative adversarial network and compressive autoencoder
Malik Qasaimeh,Alaa Abu Qtaish,Shadi Aljawarneh
DOI: https://doi.org/10.1007/s11042-024-20422-5
IF: 2.577
2024-11-16
Multimedia Tools and Applications
Abstract:Nowadays, social media applications and websites have become a crucial part of people's lives; for sharing their moments, contacting their families and friends, or even for their jobs. However, the fact that these valuable data are transferred via the Internet and open channels, which are vulnerable to attacks and espionage, requires using defense methods to improve the security of data transmission. Recently, the progress of Deep Learning (DL) inspired researchers to use it with security methods such as steganography, the art of hiding secret data in unrelated content. DL and steganography combined enhanced the hiding properties, especially in the Coverless Image Steganography (CIS) methods. In this paper, we propose a Compressive Coverless Image Steganography (CCIS) model, which is a generated-based CIS, to improve the capacity and robustness of Steganography. This model uses the compressive autoencoder to compress the message, thus increasing the capacity, and uses the Generative Adversarial Network (GAN) to generate a stego image from the compressed vector, thus enhancing concealment ability, besides using a regression model and optimization method to extract the data. Another version was designed to enable sending binary messages by replacing the compressive autoencoder network with a robust mapping rule. Experiments show that the proposed model could improve the characteristics of Steganography. Furthermore, the proposed model could extract binary messages with 256 bits from attacked stego images with 99% recovery accuracy. Thus, capacity and robustness were enhanced using our model with both images and text as secret messages.
computer science, information systems, theory & methods,engineering, electrical & electronic, software engineering