Steganography With Constructing Neural Networks

This repository contains the source code for the paper: "Steganography With Constructing Neural Networks".

Overview

This project implements a neural network-based steganography method that embeds secret information directly into the parameters of a host neural network. It utilizes an encoder-decoder framework to conceal data within the layer weights and biases, while maintaining the host model's primary classification or task performance.

Prerequisites

• Python 3.x
• PyTorch
• Additional dependencies (check requirements in the utils scripts or typical data science environment)

Project Structure

• model_steganography.py: Core logic for embedding (encode) and extracting (decode) secret information.
• models/: Contains pre-trained encoder and decoder models (e.g., encoder128.pth, decoder128.pth).
• utils/: Utility functions for data handling, initialization, training, and testing.
• dataset/: Directory for datasets (MNIST, FashionMNIST, CIFAR-10).
• stego_models/: Storage for models with embedded information.
• plt/: Visualization and plotting scripts used for analysis.
• test/: Scripts for testing functionalities like fine-tuning and parameter pruning.

Core Usage

The ModelSteganography class in model_steganography.py provides the main interface:

from model_steganography import ModelSteganography

# Initialize with desired parameters
stego = ModelSteganography(init_function=your_init_func)

# Embed information
secret_raw, secret_bch = stego.encode(model)

# Extract information
ext_raw, ext_bch = stego.decode(model)

Experiment Data Download

https://pan.quark.cn/s/8226327ae2eb?pwd=DPMT

Citation

If you find this work useful for your research, please cite:

@article{xu2025steganography,
  title={Steganography with constructing neural networks},
  author={Xu, Chenyi and Huang, Lin and Qin, Chuan and Feng, Guorui and Zhang, Xinpeng},
  journal={IEEE Transactions on Circuits and Systems for Video Technology},
  year={2025},
  publisher={IEEE}
}

Disclaimer

This software is for academic and research purposes only. The authors take no responsibility for any misuse of this technology.