Image Tampering Localization Using a Dense Fully Convolutional Network

Overview

This is the implementation of the method proposed in "Image Tampering Localization Using a Dense Fully Convolutional Network" with tensorflow(1.10.0, gpu version). The aim of this repository is to achieve image tampering localization.

Network Architecture

image

Files structure of Dense-FCN-for-tampering-localization

  • Models
  • Results
  • testedImages
  • utilis
  • train_demo.py
  • denseFCN.py
  • test_withoutComputeMetrics.py

The pre-trained model path

The model trained with Dresden script dataset and fine-tuned with 56 NIST images was uploaded in Dropbox: https://www.dropbox.com/sh/0hkeenrfazob3ci/AAAa6X2hhDnj04LfAR2mSKi9a?dl=0

How to run

Test with the trained model

python3 test_withoutComputeMetrics.py

Train the model from scratch

python3 train_demo.py

Citation

If you use our code please cite:

@ARTICLE{9393396, author={P. {Zhuang} and H. {Li} and S. {Tan} and B. {Li} and J. {Huang}},
journal={IEEE Transactions on Information Forensics and Security},
title={Image Tampering Localization Using a Dense Fully Convolutional Network},
year={2021},
volume={16},
number={},
pages={2986-2999},
doi={10.1109/TIFS.2021.3070444}}