Image Steganography with Deep Learning Networks

Abstract

Currently there are many ways to achieve information hiding in images. Each image steganography method works to increase the payload capacity while hiding the secret information in the cover image in an undetectable way, and then the receiver is able to use the extraction network to perfectly recover the secret information from the stego image. In this paper, we explore how three different network structures (convolutional neural network structure, U-Net structure, and Swin Transformer structure) solve the image embedding and extraction problem. We use the same dataset to validate the three network structures and visualize the process and effectiveness of the three network structures in achieving image steganography from different experimental results, in addition to using the peak signal-to-noise ratio (PSNR) and structural similarity between images (SSIM) to measure the image quality.