Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A high-capacity linguistic steganography based on entropy-driven rank-token mapping

Jun Jiang, Weiming Zhang, Nenghai Yu, Kejiang Chen

TL;DR

RTMStega introduces an entropy-driven rank-token mapping framework to overcome low payload in generative linguistic steganography. By compressing secret messages into token probability ranks, applying a Huffman-inspired binary mapping, and using context-aware, normalized-entropy sampling with a $\beta$-bit representation, it achieves higher embedding density while preserving naturalness. Experimental results show a roughly threefold increase in payload and over a 50% reduction in processing time across multiple models and datasets, with stego text remaining on par with cover text in perplexity and resisting standard steganalysis. While promising, the work notes that formal security proofs are still open and points to adaptive entropy thresholds as a direction for improving undetectability.

Abstract

Linguistic steganography enables covert communication through embedding secret messages into innocuous texts; however, current methods face critical limitations in payload capacity and security. Traditional modification-based methods introduce detectable anomalies, while retrieval-based strategies suffer from low embedding capacity. Modern generative steganography leverages language models to generate natural stego text but struggles with limited entropy in token predictions, further constraining capacity. To address these issues, we propose an entropy-driven framework called RTMStega that integrates rank-based adaptive coding and context-aware decompression with normalized entropy. By mapping secret messages to token probability ranks and dynamically adjusting sampling via context-aware entropy-based adjustments, RTMStega achieves a balance between payload capacity and imperceptibility. Experiments across diverse datasets and models demonstrate that RTMStega triples the payload capacity of mainstream generative steganography, reduces processing time by over 50%, and maintains high text quality, offering a trustworthy solution for secure and efficient covert communication.

A high-capacity linguistic steganography based on entropy-driven rank-token mapping

TL;DR

RTMStega introduces an entropy-driven rank-token mapping framework to overcome low payload in generative linguistic steganography. By compressing secret messages into token probability ranks, applying a Huffman-inspired binary mapping, and using context-aware, normalized-entropy sampling with a -bit representation, it achieves higher embedding density while preserving naturalness. Experimental results show a roughly threefold increase in payload and over a 50% reduction in processing time across multiple models and datasets, with stego text remaining on par with cover text in perplexity and resisting standard steganalysis. While promising, the work notes that formal security proofs are still open and points to adaptive entropy thresholds as a direction for improving undetectability.

Abstract

Linguistic steganography enables covert communication through embedding secret messages into innocuous texts; however, current methods face critical limitations in payload capacity and security. Traditional modification-based methods introduce detectable anomalies, while retrieval-based strategies suffer from low embedding capacity. Modern generative steganography leverages language models to generate natural stego text but struggles with limited entropy in token predictions, further constraining capacity. To address these issues, we propose an entropy-driven framework called RTMStega that integrates rank-based adaptive coding and context-aware decompression with normalized entropy. By mapping secret messages to token probability ranks and dynamically adjusting sampling via context-aware entropy-based adjustments, RTMStega achieves a balance between payload capacity and imperceptibility. Experiments across diverse datasets and models demonstrate that RTMStega triples the payload capacity of mainstream generative steganography, reduces processing time by over 50%, and maintains high text quality, offering a trustworthy solution for secure and efficient covert communication.
Paper Structure (16 sections, 12 equations, 2 figures, 2 tables, 2 algorithms)

This paper contains 16 sections, 12 equations, 2 figures, 2 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Related Work
  3. Generative Language Model as Sampler.
  4. Generative Steganography.
  5. Methodology
  6. Overview.
  7. Message Encoding.
  8. Message Embedding.
  9. Message Extraction.
  10. Message Decoding.
  11. Experiment
  12. Experimental Setup.
  13. Main Performance.
  14. Parameter Explorement.
  15. Steganalytic Results.
  16. ...and 1 more sections

Figures (2)

  • Figure 1: The overall framework of RTMStega, consisting of four steps: message encoding, message embedding, message extraction, and message decoding.
  • Figure 2: Variation curves illustrating the relationship between stego text payload capacity and text quality as functions of $\alpha$ and $\beta$ are generated by RTMStega. Here, $\alpha$ represents the threshold for determining whether to perform rank-based sampling based on the entropy of the model's output distribution, and $\beta$ dictates the volume of secret information embedded in a single step.