DF-Net: The Digital Forensics Network for Image Forgery Detection

Digital Forensics Network (DF-Net)

This is the official model for the DF-Net (IMVIP 2023). For method details, please refer to

  @inproceedings{Fischinger2023DFNet,
      title={DF-Net: The Digital Forensics Network for Image Forgery Detection},
      author={David Fischinger and Martin Boyer},
      journal={The 25th Irish Machine Vision and Image Processing conference. (IMVIP)},
      year={2023}
  }

 

The provided zip file DFNet.zip includes:

- Documentation: README.md
- Models: The DF-Net comprising the sub-models M1 and M2 (in TensorFlow format)
- Docker-Image: Dockerfile and code to create an independent docker container to easily test the DF-Net and reproduce results on   the benchmark dataset
- Jupyter Notebook: Interactive way to test DF-Net and visualize examples
- Testdata: Copy of the benchmark dataset CASIA* (due to space restrictions only a subset of the CASIA V1 benchmark dataset is used)
*) [Dong et al., 2013] Dong, J., Wang, W., and Tan, T. (2013). Casia image tampering detection evaluation
database. In IEEE China Summit Inter. Conf. Signal Info. Proc., pages 422–426. IEEE.

 

HowTo: Test DF-Net in Docker container

Download und extract the file DFNet.zip to an environment supporting Docker, ideally in a Linux-based system with GPU support ($ = command prompt):

 

0 - Install zip tool (if not already available):

    $ sudo apt install zip

 

1 - Download DFNet.zip

    $ wget https://zenodo.org/record/8142658/files/DFNet.zip

 

2 - Unpack zip:

    $ unzip DFNet.zip

 

3 - Change Directory:

    $ cd IMVIP_Supplementary_Material

 

4 - BUILD Docker Image

    $ ./scripts/build_image.sh

 

5 - RUN Docker Container with or without GPU

    $ ./scripts/run_container.sh

    or

    $ ./scripts/run_container_NO_GPU.sh  

 

6 - Open Jupyter Notebook

In a browser, open the link that was printed in your console (should look similar to http://127.0.0.1:8888/?token=73a6fd55dacf4984f616fe60838d64e96abba2087a6cfbda)

 

On the webpage, click on the folder dfnet and then the Jupyter notebook IMVIP_Model_Evaluation.ipynb

 

7 - Execute Code

Run all code blocks (by pressing shift+enter for each block).

Examples from the CASIA benchmark dataset will be displayed on the screen.

 

Parameter Setting

To control the number of processed images (MAX_EXAMPLES) and define how many images are displayed (SHOW_NTH_RESULT), you can modify the corresponding variables at the bottom of the Jupyter page.

 

    SHOW_NTH_RESULT = 1 # defines how many results are shown: 1...show each result;  n...only show n-th result

    MAX_EXAMPLES = 20   # defines number of images to process:  np.infty...all images in folder are processed

 

Folder Structure for Supplementary Material

.

├── datasets

│   │  

│   └── benchmark

│       ├── CASIA

│       └── CASIA_GT

├── models

│   ├── model1

│   └── model2

└── scripts

 

== Licence ==

The Software is made available for academic or non-commercial purposes only.
For commercial license pricing and support pricing, please contact:

Martin Boyer
Senior Research Engineer
Center for Digital Safety & Security
AIT Austrian Institute of Technology GmbH
Giefinggasse 4 | 1210 Vienna | Austria
martin.boyer [at] ait.ac.at

 

== CITATION ==

Please cite:
"DF-Net: The Digital Forensics Network for Image Forgery Detection", David Fischinger and Martin Boyer, IMVIP -The 25th Irish Machine Vision and Image Processing conference, (2023)