Towards a reliable detection of forgeries based on demosaicing

Author: Jérémie Dentan

This repository is the implementation of a review of the paper "Demosaicing to Detect Demosaicing and Image Forgeries" written by Quentin Bammey, Rafael Grompone von Gioi, and Jean-Michel Morel, which proposes a demosaicing-based forgery detection method.

• The report by the author: http://dx.doi.org/10.13140/RG.2.2.17584.02568
• The original paper, by Quentin Bammey, Rafael Grompone von Gioi, Jean-Michel Morel: https://doi.org/10.1109/WIFS55849.2022.9975454

Overview

For an easy demonstration of the experiments made by the author for this review, we invite you to check out the two following repositories:

• For experiment 1, about the detection of demosaicing configuration: doc/configuration_detection_demonstration.ipynb
• For experiment 2, about the detection of the image forgeries: doc/forgery_detection_demonstration.ipynb

Run the full experiments

This section explains how to run the full experiments that were run by the author for the reiew.

Set up

The code of is expected to run in Python 3.9 with the dependencies of requirements.txt installed and your PYTHONPATH set to the root of the repository. To do so, execute the following line from the root (i.e. from the folder containing this README file).

pip install -r requirements.txt
export PYTHONPATH=$(pwd)

Moreover, if you want to execute locally the notebooks in the \doc folder, you should have ipykernel installed, which is not declared in requirements.txt for technical reasons linked to the possibility to execute the notebooks on Google Colab. Thus, you should run the following before running the notebooks:

pip install notebook==6.5.3

Execution

After having set up your environment, you may run the following lines to reproduce the experiments:

python src/experiments/evaluate_config_detection.py
python src/experiments/evaluate_forgery_detection.py

This will produce some results in the output directory, as well as logs in the logs directory.

Citation

When using this repository, please cite the paper it implements:

@techreport{dentan_towards_2023,
	title = {Towards a reliable detection of forgeries based on demosaicing},
	author = {Dentan, Jérémie},
	month = mar,
	year = {2023},
	doi = {10.13140/RG.2.2.17584.02568},
}

License and Disclaimer

You may use this software under the Apache 2.0 License. See LICENSE.