Foveation Improves Payload Capacity in Steganography

TL;DR

This work tackles increasing payload capacity in image steganography while preserving visual quality by combining latent representations with foveated rendering. It introduces a Metameric Foveated Rendering loss within a frozen encoder/decoder framework to embed and recover payloads via a learnable merger and retriever. The study reports a leap from 100 to 500 bits of payload capacity with high recovery accuracy, and perceptual gains (PSNR ~31.5 dB, LPIPS ~0.13), while acknowledging resolution-imposed limits and outlining directions for robustness and gaze-aware improvements. Overall, the approach enables higher-capacity, high-fidelity steganography suitable for practical, resource-constrained scenarios.

Abstract

Steganography finds its use in visual medium such as providing metadata and watermarking. With support of efficient latent representations and foveated rendering, we trained models that improve existing capacity limits from 100 to 500 bits, while achieving better accuracy of up to 1 failure bit out of 2000, at 200K test bits. Finally, we achieve a comparable visual quality of 31.47 dB PSNR and 0.13 LPIPS, showing the effectiveness of novel perceptual design in creating multi-modal latent representations in steganography.

Figures (2)

• Figure 1: Visual example of stego and payload recovery using our proposed foveated steganography and RoSteALS (Source: MetFaces karras2020training).
• Figure 2: Our proposed foveated steganography approach (Source: MetFaces karras2020training).