NiranjanVRam / Forgeryguard

Python tool for image forgery detection

Source code for: "ForgeryGuard: Web-based Image Authenticity Checker"

Getting Started

Make sure your system has python 3.6.8rc1 version. If not, you can download it from this website.

Clone the github repo using: git clone https://github.com/NiranjanVRam/Forgeryguard.git

Requirements have been mentioned at end of this file.

There are two ways of execution:

1. Through single module:

• run app.py directly from terminal or any IDE of your choice.
• Localhost opens automatically.
• Input image and click on process, wait for 3-8 minutes based on the input(not on size).

2. Through a series of program files: Suggested IDE for execution: VS Code

• Run 0_compute_visualize_similarity_graph
• Run 1_forgery_detection
• Run 2_forgery_localization

While using this method, results(images, matrices, graphs, outputs) will directly be displayed in IDE itself.

Evaluating the tool

Modules 3, 4 and 5 given below are for evaluation of the tool(ForgeryGuard).

It consumes more execution time and memory.

For 3, 4 and 5, you need to download each dataset below from its sources:

• Columbia
• Carvalho
• Korus

If any image from the dataset is not of .TIFF or .tiff format, change to it to same.

Details for all the modules of the project

Please follow the examples in jupyter notebooks in the main directory for how to use this code.

• 0_compute_visualize_similarity_graph shows how to compute the forensic similarity graphs
• 1_forgery_detection shows how to compute forgery detection scores
• 2_forgery_localization shows how to compute forgery localization prediction masks
• 3_compute_and_save_simgraph_for_benchmark_DBs computes the forensic graphs for the three tampering datasets and saves to disk, to be used in subsequent notebooks
• 4_benchmark_forgery_detection reproduces forgery detection results
• 5_benchmark_forgery_localization reproduces forgery localization results

Note: the script 3a_format_dbs.sh should be run prior to running notebooks 3-5. Alternatively you can modify the notebooks according to your own file structure.

Requirements

• Python 3.6.8
• tensorflow 1.14.0 (gpu version recommended)
• pillow
• seaborn
• tqdm
• python-igraph (available from conda-forge)
• python flask
• other requirements to be satisfied as per requirements.txt; depends on what all libraries have already been installed on your device.