| 引用本文： |
-
陈映泉,李千目,黎汇枫,吴晓聪.增强隐写载体嵌入容量的隐藏信息转化策略及其多种应用机制[J].信息安全学报,已采用 [点击复制]
- Chen Yingquan,Li Qianmu,Li HuiFeng,Wu Xiaocong.A Hidden Information Transformation Strategy for En-hancing Embedding Capacity in Steganographic Covers and Its Various Application Mechanisms[J].Journal of Cyber Security,Accept [点击复制]
|
|
| 摘要: |
| 受限于数据载体的固有属性和有限的表达维度，传统文本隐写方法在有限空间中难以实现高载荷信息的有效隐藏，常导致嵌入容量不足、语义失真或可检测性增强等问题，从而限制了有载体文本隐写技术在实际场景中的可用性与实用性。针对上述问题，本文提出一种基于文本检索模型与中国剩余定理的隐藏信息转化策略（HITS），该策略可在保证文本自然性和语义一致性的前提下显著提升嵌入容量和信息安全性。在此基础上，本文进一步结合位置编码机制与同义词替换方法，设计了一种高容量、强隐蔽、抗主动攻击的非盲提取单模态文本隐写机制。该机制能够在语义层与结构层同时实现隐写信息的精确嵌入与鲁棒恢复，有效平衡了隐蔽性、可读性与抗分析性。为了突破单模态隐写的容量与安全瓶颈，本文又将该机制扩展至多模态隐写框架，结合文本转图像生成模型与多种经典图像隐写算法（包括最低有效位法、离散余弦变换和离散小波变换），实现了跨模态语义一致性与多层信息融合。实验结果表明，所提出的单模态隐写机制在安全性、嵌入容量及抗分析能力方面均优于现有主流方法，而多模态隐写载体在保持高语义一致性的同时，展现出更强的抗检测性与广泛的应用潜力。 |
| 关键词: 有载体隐写术 嵌入容量 文本检索模型 同余转换 同义词替换 多模态隐写术。 |
| DOI： |
| 投稿时间：2025-06-11修订日期：2025-10-21 |
| 基金项目:2024年度江苏省前沿技术研发计划项目“面向 AI 算力网络的智能系统跨域多维安全技术研发”(编号：BF2024071) |
|
| A Hidden Information Transformation Strategy for En-hancing Embedding Capacity in Steganographic Covers and Its Various Application Mechanisms |
|
Chen Yingquan, Li Qianmu, Li HuiFeng, Wu Xiaocong
|
| (Nanjing University of Science and Technology) |
| Abstract: |
| Due to the inherent characteristics and limited expressive dimensions of data carriers, traditional text steganography methods encounter significant challenges in effectively concealing high-capacity information within constrained textual spaces. These limitations often lead to problems such as insufficient embedding capacity, semantic distortion, and in-creased detectability, which in turn restrict the usability, robustness, and practicality of carrier-based text steganography techniques in real-world information security applications. To address these issues, this paper proposes a Hidden Infor-mation Transformation Strategy (HITS) based on text retrieval models and the Chinese Remainder Theorem (CRT). The proposed strategy transforms and distributes hidden information within the semantic space of carrier text, significantly improving embedding capacity and information security while maintaining the naturalness, fluency, and semantic con-sistency of the generated text. Building upon this foundation, a non-blind single-modal text steganography mechanism is designed by integrating positional encoding and synonym substitution techniques. This mechanism achieves high ca-pacity, strong concealment, and resistance to active attacks, enabling precise embedding and robust recovery of hidden information at both semantic and structural levels. It effectively balances imperceptibility, readability, and anti-analysis capability, ensuring that the steganographic text remains natural and resistant to detection or manipulation. To overcome the inherent limitations of single-modal steganography in terms of embedding capacity and security, the proposed mech-anism is further extended into a multi-modal steganography framework. By combining text-to-image generative models with several classical image steganography algorithms, including the Least Significant Bit (LSB) method, Discrete Cosine Transform (DCT), and Discrete Wavelet Transform (DWT), the framework achieves cross-modal semantic consistency and multi-level information fusion, thereby further enhancing embedding performance and resilience against statistical anal-ysis. Experimental results demonstrate that the proposed single-modal mechanism outperforms existing mainstream methods in terms of security, embedding capacity, and resistance to steganalysis, while the multi-modal steganographic carriers maintain high semantic coherence, exhibit stronger resistance to detection, and demonstrate broad potential for practical applications in secure communication and data protection. |
| Key words: Cover-based steganography Embedding capacity Text retrieval model Congruence transformation Synonym substitu-tion Multi-modal steganography. |
