SteganoGAN: High Capacity Image Steganography with GANs

TL;DR

SteganoGAN introduces a GAN-based end-to-end framework for high-capacity image steganography that can embed arbitrary binary data into natural images. The architecture—comprising an Encoder, Decoder, and Critic—utilizes three encoder connectivity variants (Basic, Residual, Dense) and a novel RS-BPP metric to fairly quantify reliable payload. Experiments on Div2K and COCO show state-of-the-art payloads up to 4.4 bpp with competitive image quality and strong evasion of standard steganalysis tools, while neural steganalysis remains challenging at moderate payloads. The work provides an open-source library and a robust evaluation framework for comparing deep-learning steganography against traditional methods.

Abstract

Image steganography is a procedure for hiding messages inside pictures. While other techniques such as cryptography aim to prevent adversaries from reading the secret message, steganography aims to hide the presence of the message itself. In this paper, we propose a novel technique for hiding arbitrary binary data in images using generative adversarial networks which allow us to optimize the perceptual quality of the images produced by our model. We show that our approach achieves state-of-the-art payloads of 4.4 bits per pixel, evades detection by steganalysis tools, and is effective on images from multiple datasets. To enable fair comparisons, we have released an open source library that is available online at https://github.com/DAI-Lab/SteganoGAN.

Figures (6)

• Figure 1: A randomly selected cover image (left) and the corresponding steganographic images generated by steganogan at approximately 1, 2, 3, and 4 bits per pixel.
• Figure 2: (a) The model architecture with the Encoder, Decoder, and Critic. The blank rectangle representing the Encoder can be any of the following: (b) Basic encoder, (c) Residual encoder and (d) Dense encoder. The trapezoids represent convolutional blocks, two or more arrows merging represent concatenation operations, and the curly bracket represents a batching operation.
• Figure 3: Randomly selected pairs of cover (left) and steganographic (right) images from the COCO dataset which embeds random binary data at the maximum payload of 4.4 bits-per-pixel.
• Figure 4: A randomly selected pair of cover (left) and steganographic (right) images and the differences between them. The top row shows the output from a simple least-significant-bit steganography algorithm survey while the bottom row shows the output from steganogan with 4.4 bpp. Note that steganogan is able to adapt to the image content.
• Figure 5: The receiver operating characteristic (ROC) curve produced by the StegExpose library for a set of 1000 steganographic images generated using the Dense architecture with a data depth of 6. The StegExpose library includes multiple steganalysis tools including $\mathtt{Sample Pairs}$samplepairs, $\mathtt{RS Analysis}$rsanalysis, $\mathtt{Chi Squared Attack}$chisquare, and $\mathtt{Primary Sets}$primarysets. The tool achieves an auROC of 0.59.