StegOT: Trade-offs in Steganography via Optimal Transport

TL;DR

StegOT addresses the information balance challenge in image steganography by integrating an autoencoder with a Multi-Channel Optimal Transport module to transform a multi‑peak latent distribution into a single-peak form. This redistribution enables simultaneous preservation of cover information and recovery of the secret, improving both stego and recovered image quality. Empirical results on DIV2K, COCO, and ImageNet show superior PSNR and lower perceptual distance (LPIPS) compared with prior methods, along with robustness under distortions. The approach provides a theoretically motivated and practically effective framework for more reliable steganography with higher fidelity and resilience.

Abstract

Image hiding is often referred to as steganography, which aims to hide a secret image in a cover image of the same resolution. Many steganography models are based on genera-tive adversarial networks (GANs) and variational autoencoders (VAEs). However, most existing models suffer from mode collapse. Mode collapse will lead to an information imbalance between the cover and secret images in the stego image and further affect the subsequent extraction. To address these challenges, this paper proposes StegOT, an autoencoder-based steganography model incorporating optimal transport theory. We designed the multiple channel optimal transport (MCOT) module to transform the feature distribution, which exhibits multiple peaks, into a single peak to achieve the trade-off of information. Experiments demonstrate that we not only achieve a trade-off between the cover and secret images but also enhance the quality of both the stego and recovery images. The source code will be released on https://github.com/Rss1124/StegOT.

Figures (6)

• Figure 1: In GAN-based or VAE-based steganographic models, the decoder is more inclined to sample the information of Cover image, which leads to the imbalance between Cover information and Secret information in Stego images.
• Figure 2: The framework of StegOT. In the hiding network, MCOT transforms a multi-peak feature distribution into a single-peak distribution. In this way, MCOT achieves a balance between cover information and secret information in the stego image. To protect the watermark image from leaking out, we use the optimal transmission map T generated by MCOT as the key to start the Reval Network.
• Figure 3: Visualization comparison of different Steganography methods on COCO. The higher the value of PSNR and SSIM, the better the model performance.
• Figure 4: The models utilizing MCOT (solid lines in the figure) exhibit higher PSNR values across all datasets compared to models without MCOT (dashed lines). Similarly, the SSIM values for models using MCOT are higher across all datasets than those for models not using MCOT.
• Figure 5: The gray histogram comparison between cover/secret image and stego/recovery image.