GAN-based Coverless Multi-Image Steganography with RGB Secret Images

Abstract

Abstract Steganography emerges as a powerful method for secure data transfer in the digital age when information security is becoming increasingly important. Advances in this sector have opened the way for image steganography—a technique that involves embedding complete images within another, in a manner imperceptible to unsuspecting observers, assuring the communication’s invisibility. Despite breakthroughs, existing multi-image steganography models have limits that require the creation of more advanced solutions. These solutions must not only improve image quality but also improve security measures. Comprehensive research into cutting-edge techniques has revealed that Generative Adversarial Networks (GANs) significantly improve the performance of image steganography systems by increasing their embedding capacity, improving recovery accuracy, and fortifying their security protocols. As a result, the primary goal of this research is to investigate the potential of GAN-based image steganography models for advancing the multi-image steganography area, especially using a coverless technique with RGB secrete images. The study concludes by proposing a novel coverless multi-image steganography model that leverages GANs to seamlessly embed and extract multiple secret images with minimal loss. The model’s performance was rigorously evaluated using industry-standard metrics, focusing on image reproduction accuracy as measured by RMSE, MSE, PSNR, and SSIM, which were 22.50, 546.25, 21.39, and 0.66, respectively. Additionally, the model demonstrated a hiding capacity of 48 bits per pixel. Compared to traditional image steganography methods, the proposed approach not only achieves notable improvements in concealment and accuracy but also significantly enhances embedding capacity. This study advances the field of image steganography by exploring the potential of coverless multiimage steganography through the use of GANs.