-
CASIA 2.0 groundtruth dataset is avaiable at: https://github.com/namtpham/casia2groundtruth
-
Please notice that the authors made mistakes in naming the files. If you already downloaded the dataset, I recommend you to rename the tampered images using the commands in the excel file. Otherwise, you can use the modified version of dataset.
-
Since the owners' server is no longer avaiable, I upload the dataset to my drive, including original dataset and the modified version (renaming dozens of images in tampering folder). Please visit one of the following links to download (~90 MB):
-
Google Drive: https://bit.ly/37TuHRx
-
One Drive: https://bit.ly/3nXyYJw
Due to the lack of manual file, I write up here the naming convention:
Au_ani_0001.jpg
Au: Authentic
ani: animal category
Other categories: arc (architecture), art, cha (characters), nat (nature), pla (plants), sec, txt (texture)
a. Spliced image
Sp_D_CND_A_pla0005_pla0023_0281.jpg
- Sp: Splicing
- D: Different (means the tampered region was copied from the different image)
- Next 4 letters stand for the techniques they used to create the images. Unfortunately, I don't remember exactly.
- pla0005: the source image
- pla0023: the target image
- 0281: tampered image ID
b. Copy-move images
Sp_S_CND_A_pla0016_pla0016_0196.jpg
- Sp: Tampering
- S: Same (means the tampered region was copied from the same image)
- And the rest is similar to case a.
If you use the groundtruth dataset for a scientific publication, please cite the following paper (https://res.mdpi.com/symmetry/symmetry-11-00083/article_deploy/symmetry-11-00083.pdf):
@article{pham2019hybrid,
title={Hybrid Image-Retrieval Method for Image-Splicing Validation},
author={Pham, Nam Thanh and Lee, Jong-Weon and Kwon, Goo-Rak and Park, Chun-Su},
journal={Symmetry},
volume={11},
number={1},
pages={83},
year={2019},
publisher={Multidisciplinary Digital Publishing Institute}
}