Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

DTAMS: High-Capacity Generative Steganography via Dynamic Multi-Timestep Selection and Adaptive Deviation Mapping in Latent Diffusion

Yuhao Xue, Jiuan Zhou, Yu Cheng, Zhaoxia Yin

TL;DR

DTAMS tackles the challenge of high-capacity generative steganography by integrating dynamic multi-timestep embedding, global interval deviation mapping, and multi-dimensional deviation compensation into a diffusion-based framework. It preselects embedding timesteps via a transition-cost model, maps modifications to pixel-intervals to constrain deviations, and jointly regularizes pixel, latent, and semantic representations to enhance robustness and anti-steganalysis performance. Experiments show embedding capacities up to 12 bpp with extraction accuracy above 99% and strong perceptual quality, while reducing average extraction errors by approximately 59% over state-of-the-art methods under common distortions. The approach offers practical, robust, high-capacity steganography for diffusion-generated imagery with effective defenses against standard steganalysis attacks.

Abstract

With the rapid development of AIGC technologies, generative image steganography has attracted increasing attention due to its high imperceptibility and flexibility. However, existing generative steganography methods often maintain acceptable security and robustness only at relatively low embedding rates, severely limiting the practical applicability of steganographic systems. To address this issue, we propose a novel DTAMS framework that achieves high embedding rates while ensuring strong robustness and security. Specifically, a dynamic multi-timestep adaptive embedding mechanism is constructed based on transition-cost modeling in diffusion models, enabling automatic selection of optimal embedding timesteps to improve embedding rates while preserving overall performance. Meanwhile, we propose a global sub-interval mapping strategy that jointly considers mapping errors and the frequency distribution of secret information, converting point-wise perturbations into interval-level statistical mappings to suppress error accumulation and distribution drift during multi-step diffusion processes. Furthermore, a multi-dimensional joint constraint mechanism is introduced to mitigate distortions caused by repeated latent-pixel transformations by jointly regularizing embedding errors at the pixel, latent, and semantic levels. Experiments demonstrate that the proposed method achieves an embedding rate of 12 bpp while maintaining excellent security and robustness. Across all evaluated conditions, DTAMS reduces the average extraction error rate by 59.39%, representing a significant improvement over SOTA methods.

DTAMS: High-Capacity Generative Steganography via Dynamic Multi-Timestep Selection and Adaptive Deviation Mapping in Latent Diffusion

TL;DR

DTAMS tackles the challenge of high-capacity generative steganography by integrating dynamic multi-timestep embedding, global interval deviation mapping, and multi-dimensional deviation compensation into a diffusion-based framework. It preselects embedding timesteps via a transition-cost model, maps modifications to pixel-intervals to constrain deviations, and jointly regularizes pixel, latent, and semantic representations to enhance robustness and anti-steganalysis performance. Experiments show embedding capacities up to 12 bpp with extraction accuracy above 99% and strong perceptual quality, while reducing average extraction errors by approximately 59% over state-of-the-art methods under common distortions. The approach offers practical, robust, high-capacity steganography for diffusion-generated imagery with effective defenses against standard steganalysis attacks.

Abstract

With the rapid development of AIGC technologies, generative image steganography has attracted increasing attention due to its high imperceptibility and flexibility. However, existing generative steganography methods often maintain acceptable security and robustness only at relatively low embedding rates, severely limiting the practical applicability of steganographic systems. To address this issue, we propose a novel DTAMS framework that achieves high embedding rates while ensuring strong robustness and security. Specifically, a dynamic multi-timestep adaptive embedding mechanism is constructed based on transition-cost modeling in diffusion models, enabling automatic selection of optimal embedding timesteps to improve embedding rates while preserving overall performance. Meanwhile, we propose a global sub-interval mapping strategy that jointly considers mapping errors and the frequency distribution of secret information, converting point-wise perturbations into interval-level statistical mappings to suppress error accumulation and distribution drift during multi-step diffusion processes. Furthermore, a multi-dimensional joint constraint mechanism is introduced to mitigate distortions caused by repeated latent-pixel transformations by jointly regularizing embedding errors at the pixel, latent, and semantic levels. Experiments demonstrate that the proposed method achieves an embedding rate of 12 bpp while maintaining excellent security and robustness. Across all evaluated conditions, DTAMS reduces the average extraction error rate by 59.39%, representing a significant improvement over SOTA methods.
Paper Structure (16 sections, 11 equations, 5 figures, 6 tables, 1 algorithm)

This paper contains 16 sections, 11 equations, 5 figures, 6 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. Diffusion Models
  4. Generative Image Steganography
  5. Methodology
  6. Framework Overview and Processing Pipeline
  7. Dynamic Multi-Timestep Adaptive Embedding via Preprocessing-Based Selection
  8. Sub-Interval Partitioning Based on Adaptive Deviation Mapping
  9. Multi-Dimensional Deviation Compensation
  10. Experiments
  11. Experimental Setup
  12. Flexibility of Steganographic Capacity
  13. Practicality
  14. Robustness
  15. Security
  16. ...and 1 more sections

Figures (5)

  • Figure 1: The pipeline of DTMAS.
  • Figure 2: The Framework of DTAMS.
  • Figure 3: Multi-Timestep Adaptive Embedding and Selection
  • Figure 4: The PGD full-image interval optimization highlights the combined effect of smoothness and reconstruction losses.
  • Figure 5: DTAMS generates stego images conditioned on input text, producing personalized images for different scenes and multiple variations for the same scene.