Recod.ai Scientific Image Integrity Library

A collection of algorithms to synthetically create scientific images for forensics and integrity analysis.

An in-depth explanation of each algorithm and dataset is described in our research work: Benchmarking Scientific Image Forgery Detectors

Library

The library implements the most type of image tampering functions.

1. Image Duplication
2. Retouching
3. Cleaning

This notebook explains how to apply each type of forgery in a scientific image.

The library also mimics the behavior of images placed in scientific documents, such as compound figures -- with indicative letters and graphs.

There are two possible types of forgeries for compound figures:

1. Intra-panel (forgeries that are isolated within a single panel from the compound figure)

   Notebook explaining each type of implemented forgery

2. Inter-panel (forgeries that involve more than one figure panel):

   Notebook explaining each type of implemented forgery

Requirements:

To run the notebooks, make sure to install python3.8 and the modules included in the requirements.txt.

Recod.ai Scientific Image Integrity Dataset

Using the implemented library, we created a synthetic dataset dedicated to forensics purposes and scientific integrity.

RSIID Dataset:

Train set

Test set

Source figures and compound figure templates used to create the tampering dataset:

Source Figures

Templates

Dataset Organization

Both train and test sets have simple and compound figures, organized with the following schematic:

Simple Images

Compound figure

Citation

The dataset is distributed under the Attribution 4.0 International (CC BY 4.0) license.

If you use any content from this repository, please cite:

 @article{cardenuto_2022, 
 title={Benchmarking scientific image forgery detectors},
 volume={28}, DOI={10.1007/s11948-022-00391-4},
 number={4},
 journal={Science and Engineering Ethics},
 author={Cardenuto, João P. and Rocha, Anderson}, year={2022}
 }