CardiffVisualComputing/CSVTED

Dataset Card for COMSATS Structured Video Tampering Evaluation Dataset (CSVTED)

<!-- Provide a quick summary of the dataset. -->

The COMSATS Structured Video Tampering Evaluation Dataset (CSVTED) provides a three-level dataset structured by tampering quality and video complexity, designed to support research in video forensic analysis.

Dataset Details

The dataset consists of 1047 videos: 133 original + 914 tampered in various formats (avi, mp4, mov). These were captured using multiple cameras and lighting conditions (morning, noon, evening, night, fog). We included various tampering types, both spatial and temporal:

• —Frame duplication
• —Deletion
• —Insertion
• —Copy-move
• —Splicing
• —Event-Object-Person (EOP) based tampering

We used SSIM and Optical Flow to select optimal positions for duplication, insertion, or deletion. This helps maintain visual consistency and avoids abrupt changes, especially by matching object motion direction.

Dataset Structure

The dataset consists of zipped video files in either avi, mp4 or mov formats.

From the original untampered videos:

• —Original.zip (133 videos)

7 sets of manipulated videos were derived:

• —EOP-Frame-Deletion.zip (20 videos)
• —EOP-Frame-Duplication.zip (20 videos)
• —EOP-Frame-Insertion.zip (20 videos)
• —Frame-Deletion.zip (260 videos)
• —Frame-Duplication.zip (291 videos)
• —Frame-Insertion.zip (293 videos)
• —Spatial-Tampering.zip (10 videos)

Filename Structure

The information regarding the number of tampered frames and the location of tampering is embedded in the filename/title of each tampered video. For example, consider the following tampered video: 010_Del_021_030_Shop Banners

The components of the filename represent the following:

• —10 represents the number of frames deleted.
• —Del represents the type of tampering, i.e., Frame Deletion.
• —021 represents the location of the tampering (Localization).

Thus, frames 021 through 030, including both the 21st and 30th frames, were deleted from the original/untampered video to generate the tampered video.

Similarly if the title of video is: 010_Ins_017_026_Slides

The components of the filename represent the following:

• —010 represents the number of frames inserted.
• —Ins represents the type of tampering, i.e., Frame Insertion.
• —017 represents the starting location of the insertion.
• —026 represents the ending location of the insertion.

Therefore, the insertion operation starts at the 17th frame and ends at 26th frame.

Citation [optional]

BibTeX:

@article{CSVTED,
  title={{COMSATS} Structured Video Tampering Evaluation Dataset {(CSVTED)}},
  author={N. Akhtar and M. Saddique and P.L. Rosin and X. Sun and M. Hussain and Z. Habib},
  journal = {Machine Vision and Applications},
  volume = {36},
  number = {86},
  year = {2025}
}

Dataset Card Contact

Paul Rosin\ Cardiff University\ rosinpl@cardiff.ac.uk