Forensic Similarity Graph

The following is a non-official implementation of the work "Exposing Fake Images with Forensic Similarity Graphs" by Owen Mayer and Matthew C. Stamm, Deparment of Electrical and Computer Engineering Drexel University - Philadelphia, PA, USA. Part of the code is based on the forensic similarity code base.

Prerequisites

• python 3

• python packages:

  • tensorflow 1.14.0
  • numpy 1.16.4
  • tqdm
  • igraph 0.8.2

• optional recommended python packages:

  • jupyter notebook (for working with example scripts)
  • matplotlib (for loading images in the examples)
  • cairocffi 1.1.0 (for generating plot examples)
  • pillow (for loading JPEG images)
  • scikit-learn 0.22.1 (for metrics calculation)

Getting Started

Please see the jupyter notebook examples to get started.

The "src" folder contains all the code that you need to run the project. To begin, create a "data" folder in the main directory of this project and paste your dataset. Then, you can run the project and change some basic configuration inside the "test.py" file.

Cite the original paper

For more information regarding this work, please refear to the original paper.

bibtex:

@article{mayer2019forensicgraph,
  title={Exposing Fake Images with Forensic Similarity Graphs},
  author={Mayer, Owen and Stamm, Matthew C},
  journal={arXiv preprint arXiv:1912.02861},
  year={2019}
}