Comprint is an image forgery detection and localization method that utilizes compression fingerprints.

These files were created by IDLab-MEDIA, Ghent University - imec, in collaboration with the Image Processing Research Group of the University Federico II of Naples (GRIP-UNINA).

All rights reserved.

This software should be used, reproduced and modified only for informational and nonprofit purposes.

By downloading and/or using any of these files, you implicitly agree to all the terms of the license, as specified in the document LICENSE.txt (included in this package).

IDLab-MEDIA: https://media.idlab.ugent.be/

GRIP-UNINA: https://www.grip.unina.it/

The code requires Python 3.X and was built with Tensorflow 2.9.1.

Install the requested libraries using:

pip install -r code/requirements.txt

First, download the training and validation data, and place it in data/train and data/validation, respectively. Downloadlinks can be found in data/downloadlinks_train_and_validation.txt.

Training settings can be changed with the corresponding variables in code/train_network.py and train_network_siamese.py.

Then the shell scripts in the main folder start the training. For training the pre-trained network that estimates JPEG artifacts:

bash run-training.sh

For training the siamese network that extracts the comprint:

bash run-training-siamese.sh

The Jupyter notebook code/get_comprint_heatmap.ipynb gives an example on how to extract the comprint and heatmap. By changing the filename / path, you can extract the comprint from other images under investigation. Our trained models are included in the models folder.

More information can be found on our website.

The paper can be downloaded on arXiv.

Alternatively, the conference presentation was recorded and uploaded on YouTube, and can be watched here.

Click on the image to go to the YouTube video

This work was presented in the Workshop on MultiMedia FORensics in the WILD (MMFORWILD) 2022, held in conjunction with the International Conference on Pattern Recognition (ICPR) 2022.

@InProceedings{mareen2022comprint,
  author="Mareen, Hannes and Vanden Bussche, Dante and Guillaro, Fabrizio and Cozzolino, Davide and Van Wallendael, Glenn and Lambert, Peter and Verdoliva, Luisa",
  editor="Rousseau, Jean-Jacques and Kapralos, Bill",
  title="Comprint: Image Forgery Detection and Localization Using Compression Fingerprints",
  booktitle="Pattern Recognition, Computer Vision, and Image Processing. ICPR 2022 International Workshops and Challenges",
  year="2023",
  publisher="Springer Nature Switzerland",
  address="Cham",
  pages="281--299",
  doi="10.1007/978-3-031-37742-6_23",
}