Skip to content

[NeurIPS 2023] Official PyTorch implementation for the paper "CRoSS: Diffusion Model Makes Controllable, Robust and Secure Image Steganography"

Notifications You must be signed in to change notification settings

vvictoryuki/CRoSS

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

26 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

CRoSS (NeurIPS 2023)

CRoSS: Diffusion Model Makes Controllable, Robust and Secure Image Steganography

Jiwen Yu1, Xuanyu Zhang1, Youmin Xu1, 2, Jian Zhang1

1 Peking University, 2 PCL

arXiv

News

  • News (2024-06-12): 🎉🎉🎉 Our Stego260 has been released here. The image-prompt pair can be downloaded here.
  • News (2023-09-28): 🎉🎉🎉 Our code has been released!
  • News (2023-09-22): 🎉🎉🎉 Congratulations on CRoSS being accepted by NeurIPS 2023! Our open-source project is making progress, stay tuned for updates!

Setup

Try the command:

pip install -r requirements.txt

Quick Run

python demo.py --image_path ./asserts/1.png --private_key "Effiel tower" --public_key "a tree" --save_path ./output --num_steps 50

Feel free to try it on your own images and keys. More sampling steps usually lead to better results.

Introduction

Inspired by recent developments in diffusion models, we propose a novel image steganography framework named Controllable, Robust, and Secure Image Steganography (CRoSS). This framework offers significant advantages in controllability over container images, robustness against complex degradation during transmission of container images, and enhanced security compared to cover-based image steganography methods. Importantly, these benefits are achieved without requiring additional training.

Results about Robustness

Following are visual comparisons of our CRoSS and other methods under two real-world degradations, namely WeChat and Shoot. Obviously, our method can reconstruct the content of secret images, while other methods exhibit significant color distortion or have completely failed.

Details and more results can be found in our paper.

Results about Security

Following are deep steganalysis results by the latest SID. As the number of leaked samples increases, methods whose detection accuracy curves grow more slowly and approach $50%$ exhibit higher security. The right is the recall curve of different methods under the StegExpose detector. The closer the area enclosed by the curve and the coordinate axis is to 0.5, the closer the method is to the ideal evasion of the detector.

Details and more results can be found in our paper.

Citation

If this work is helpful for your research, please consider citing the following BibTeX entry.

@article{yu2023cross,
title={CRoSS: Diffusion Model Makes Controllable, Robust and Secure Image Steganography},
author={Yu, Jiwen and Zhang, Xuanyu and Xu, Youmin and Zhang, Jian},
journal={Advances in Neural Information Processing Systems (NeurIPS)},
year={2023}
}

About

[NeurIPS 2023] Official PyTorch implementation for the paper "CRoSS: Diffusion Model Makes Controllable, Robust and Secure Image Steganography"

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages