We are making freely available for research purposes the following dataset:
Concurrent Photo Sequence Dataset
The dataset has been collected during a public event and it is composed of 4 photo sequences: sequence User 1, User 2, User 3, User 4.
|The 4 photo sequences have been collected concurrently by 4 different photographers using different cameras and setting. Each picture is saved as a jpg image with different resolutions depending on the camera setting adopted.|
More information on the dataset and our results in “Concurrent Photo Sequence Organization”, MTAP 2012, (pdf) You may download the dataset as a zip file from this link. This dataset is for non-commercial research purposes (such as academic research) only. Any other use is forbidden. The images are not allowed to be redistributed (do not pass copies of any part of this collection to others, or post any images on the Internet).
Copy-Move Forgery Dataset
The Dataset is made of medium sized images (almost all 1000×700 or 700×1000) and it is subdivided into several datasets (D0, D1, D2). The first dataset D0 is made of 50 not compressed images with simply translated copies. For the other two groups of images (D1, D2) we selected 20 not compressed images, representing simple scenes (single object, simple background), rather than complex scenes, as we are interested in studying primarily the robustness against some specific attacks. You may download the dataset files from this link. More information and details about the dataset and our results are in “Copy-Move Forgery Detection by Matching Triangles of Keypoints” – IEEE Transaction on Information Forensics and Security, 2015 from THIS LINK (pdf Copy) DOWNLOAD DATASET Details :
- D1 has been created by copy-pasting objects after rotation
- D2 applying scaling to the copies.Each dataset has been further subdivided into subsets.
- The first subset D1.1 has been created applying to the copies 11 different types of rotation around the angle zero in the range of [-25°, 25°] with step 5°.
- The second subset D1.2 is created by rotating the copies by 12 different angles in the range of [0°, 360°[ with a step of 30°.
- The third subset D1.3 is built by rotating the copies by 11 different angles in the range of [-5°, 5°] with a step of 1°.
- D1 is then composed by 680 images (with some repetitions) .
- The subset D2.1 is obtained by scaling the copies by 8 different scaling factors in the range of [0.25, 2] with step 0.25.
- In D2.2 copies are scaled by 11 scaling factors in the range of [0.75, 1.25] with step 0.05. D2 is then composed of 380 images (with some intersections).
- The subset D3 is made of 50 original images without tampering
Tampered Video Dataset
The Dataset includes 160 tampered videos, from 6 different original videos. Tampered videos are made selecting an object in a frame of a video, and tracking the object for a certain number of frames. The copied object is cloned into another part of the same video, after possible transformations. The dataset includes:
- The 6 original videos.
- The 160 tampered videos, divided per original video and subdivided per type of transformation.
For each tampered video we included also two Matlab files which report information about the tracking and the cloning process (see the README file included into the zip for further details). This information can be used to reconstruct the cloning process of the object (starting from the selection step, to the transformation and to the pasting process), and to measure the results of the Video Forensics methods. You may download the dataset files from this link. More information and details about the dataset and our results are in: ARDIZZONE E., MAZZOLA G., (2015). “A Tool to Support the Creation of Datasets of Tampered Videos”, International Conference on Image Analysis and Processing, ICIAP 2015, pp. 665-675, DOI 10.1007/978-3-319-23234-8_61. that can be dowloaded from THIS LINK (pdf Copy)