ReLoc: A Restoration-Assisted Framework for Robust Image Tampering Localization

Overview

This is the implementation of the method proposed in "ReLoc: A Restoration-Assisted Framework for Robust Image Tampering Localization" with Pytorch(1.9.0 + cu102). The aim of this repository is to achieve robust image tampering localization.

Network Architecture!

Files structure of ReLoc

codes
- models: codes of SCSEUnet ^[1]
- MVSS_net: codes of MVSSNet ^[2]
- denseFCN.py: code of DFCN ^[3]
- SCUNet_main: codes of SCUNet ^[4]
- metrics.py: code for computing the localization performance.
- test.py: the testing script.
- train.py: the training script.
- configs.py: the config of training ReLoc.
checkpoints: the weights of ReLoc equipped with 3 localization modules (i.e., DFCN, SCSEUnet, and MVSSNet) trained on DEFACTO dataset. You can download these files from Baidu Yun (Code: e5ww)

How to run

Train the ReLoc model

1. cd ./codes

2. Modify the training config of ReLoc in configs.py

3. python train.py

Test the ReLoc model

1. python test.py

Acknowledgments

The tampering localization methods and restoration method used in this paper can find in the following links:

SCSE-Unet ^[1]: paper and codes
MVSSNet ^[2]: paper and codes
DFCN ^[3]: paper and codes
SCUnet ^[4]: paper and codes

Citation

If you use our code please cite:

@ARTICLE{ReLoc,

title={ReLoc: A Restoration-Assisted Framework for Robust Image Tampering Localization},

author={Zhuang, Peiyu and Li, Haodong and Yang, Rui and Huang, Jiwu},

journal={IEEE Transactions on Information Forensics and Security},

year={2023},

volume={18},

pages={5243-5257}}

shivinn/ReLoc