Enhanced Pixel Privacy: Leveraging Deep Learning for Advanced Image Steganography

Digital image steganography is the practice of safely, undetectably, and retrievably obfuscating data in images. In our increasingly digital world, data breaches are becoming more common, posing a serious threat to data security, including that of digital images shared or stored online. Various steganographic methods have been proposed to address these concerns, largely involving the concealment of a secret image within a host image. Traditional methods, however, have limitations, including restricted data embedding capacity and lower quality of the reconstructed image. This research study proposes and implements a novel model, the Enhanced Pixel Privacy GAN (EPPGAN), which combines a multi-attention-based Swin Transformer with a Cycle GAN training approach. A detailed comparison of the performance of various CNN architectures in hiding a secret image within a host image is performed. The proposed technique outperforms the other existing techniques in terms of image quality and embedding capability.