Code-Level Cost Function Generation for Spatial Image Steganography Using RAG-Enhanced Large Language Models

2026-07-07

AI Summary

This work proposes a retrieval-augmented generation (RAG)-based self-evolving large language model (LLM) system to address limitations in existing adaptive steganography cost function design, which either relies on manual parameter tuning or suffers from the poor interpretability of deep learning approaches. The proposed method translates generation objectives into queries via code semantic signatures and retrieves guiding information from both static literature and dynamic experiential knowledge bases to drive the automatic synthesis of compliant, interpretable cost functions. A feedback mechanism continuously refines the knowledge base to enhance performance. Experimental results demonstrate that the approach achieves superior steganographic security on BOSSBase and BOWS2 compared to current automated design methods, with a 46.3% increase in code execution rate and a 26.1% reduction in search cost.

Abstract

Designing cost functions of adaptive steganography traditionally requires extensive manual tuning, while deep learning methods lack interpretability. Although large language models (LLMs) offer an automated alternative via evolutionary generation, they often violate domain specific mathematical constraints due to a lack of explicit domain knowledge. To address this problem, we propose a novel evolutionary system focused on exploiting Retrieval-Augmented Generation (RAG) enhanced LLMs for the automatic code-level generation of spatial steganography cost functions. This system incorporates a core Self Evolving RAG (SE-RAG) module, wherein a Code Semantic Signature (CSS) translates procedural code into aligned queries, retrieving explicit guidance from static literature and dynamic experience knowledge bases to steer the LLM generation process. A dedicated feedback mechanism then continuously refines the dynamic knowledge base with successful optimization strategies. Extensive experiments on the BOSSBase and BOWS2 datasets demonstrate that the proposed framework consistently achieves higher steganographic security than existing automatically designed methods, and increases the average code execution rate by 46.3% while reducing the search cost by 26.1%, thereby highlighting the effectiveness, efficiency, and potential of combining LLMs with domain-specific knowledge in the field of automatic steganographic algorithm generation.

Problem

• steganography
• cost function
• large language models
• domain knowledge
• RAG

Innovation

• Retrieval-Augmented Generation
• Large Language Models
• Steganographic Cost Function
• Self-Evolving RAG
• Code Semantic Signature