Breaking the Resolution Barrier: Arbitrary-resolution Deep Image Steganography Framework

TL;DR

ARDIS tackles the fixed-resolution constraint in deep image steganography by introducing a reference-guided, continuous-reconstruction framework. It decouples secret content into a resolution-aligned global basis and a resolution-agnostic high-frequency latent (FDA), uses a latent-guided implicit reconstructor (LGIR) for deterministic, pixel-level recovery at arbitrary resolutions, and applies an implicit resolution coding (IRC) strategy to enable blind resolution decoding. The approach yields superior invisibility and cross-resolution recovery fidelity compared with state-of-the-art methods, including zero-error blind recovery of secret resolutions. This has practical implications for covert communications and security applications where resolution metadata cannot be transmitted or synchronized.

Abstract

Deep image steganography (DIS) has achieved significant results in capacity and invisibility. However, current paradigms enforce the secret image to maintain the same resolution as the cover image during hiding and revealing. This leads to two challenges: secret images with inconsistent resolutions must undergo resampling beforehand which results in detail loss during recovery, and the secret image cannot be recovered to its original resolution when the resolution value is unknown. To address these, we propose ARDIS, the first Arbitrary Resolution DIS framework, which shifts the paradigm from discrete mapping to reference-guided continuous signal reconstruction. Specifically, to minimize the detail loss caused by resolution mismatch, we first design a Frequency Decoupling Architecture in hiding stage. It disentangles the secret into a resolution-aligned global basis and a resolution-agnostic high-frequency latent to hide in a fixed-resolution cover. Second, for recovery, we propose a Latent-Guided Implicit Reconstructor to perform deterministic restoration. The recovered detail latent code modulates a continuous implicit function to accurately query and render high-frequency residuals onto the recovered global basis, ensuring faithful restoration of original details. Furthermore, to achieve blind recovery, we introduce an Implicit Resolution Coding strategy. By transforming discrete resolution values into dense feature maps and hiding them in the redundant space of the feature domain, the reconstructor can correctly decode the secret's resolution directly from the steganographic representation. Experimental results demonstrate that ARDIS significantly outperforms state-of-the-art methods in both invisibility and cross-resolution recovery fidelity.

Figures (6)

• Figure 1: Overview of our ARDIS compared with existing methods. (a) Existing methods: The sender must manually resample the secret image to the same resolution as the cover image to hide it, and the recovered secret image will look strange due to the different resolution (e.g., different aspect ratio). (b) Our ARDIS enables arbitrary-resolution hiding and high-fidelity blind restoration.
• Figure 2: Overview of the proposed ARDIS, which supports hiding and revealing secret images at arbitrary resolutions.
• Figure 3: Visual comparisons of our ARDIS with leading deep image steganography methods for stego and recovering secret images in arbitrary-resolution hiding scenarios. The recovery of secret images presents two scenarios. (1) Without explicitly transmitting resolution metadata, no comparison methods can recover the secret image at the original resolution. (2) Explicitly transmitting resolution metadata, the comparison methods relies on resampling for recovery, which results in the loss of many details.
• Figure 4: The visual comparison of ARDIS and the diffusion-based DIS method on the Stego260 dataset. The diffusion-based DIS method requires no cover and simply follows the prompts to generate the corresponding stego image. The settings for prompts 1 and 2 completely follow the design of the CRoSS and Diffstega methods.
• Figure 5: Steganalysis accuracy by SRNet. The fact that the curve remains close to 50$\%$ despite the increasing number of leaked samples demonstrates the high security of the method.