Fragmented Image Repository at LSU GVC Lab

About the Dataset

This repository is our collection of generated or scanned fragmented images. This repository could serve as a testing benchmark for image reassembly algorithms. This Image Reassembly Repository is a comprehensive dataset, currently containing 450 sets of image fragments. 400 of them are digitally generated; 40 are hand-torn; and 10 are from scanned Jigsaw puzzles. Each fragment set consists of 12 ~ 500 fragments. Some fragment set consists of fragments from more than one source image. The groundtruth information is included with the data.

Image Reassembly Demos

Fragmented Image Reassembly

The data is available upon request to xinli@lsu.edu.

Digitally Generated Fragments (Clean Fragments with Groundtruth Transformations from the Original Complete Image): 50 Sets

Scene_NewYorkCity (36 pieces)

Hawaii And Vinice City (50 pieces from 2 images)

3 Postcards (48 pieces from 3 images)

Scan of Hand-torn Fragments (Scan with Noise, with the Complete Image): 30 Sets

Torn_Fox (16 pieces)

Torn_Minion (17 pieces)

Torn_Tiger_2 (23 pieces)

Torn_Dog (28 pieces)

Torn_Tiger_1 (70 pieces)

Scan of Jigsaw Puzzles (Scan with Noise, with the Complete Image): 10 Sets

Puzzle_Stegosaurus (12 pieces)

Puzzle_Frozen (42 pieces)

The data is available upon request to xinli@lsu.edu.

Referencing the Dataset in your work

Here are the Bibtex snippets for citing the Dataset in your work.

        @article{Zhang14GMOD,
          author    = {Kang Zhang and Xin Li},
          title     = {A graph-based optimization algorithm for fragmented image reassembly},
          journal   = {Graphical Models},
          year      = {2014},
          volume    = {76},
          number    = {5},
          pages     = {484--495}
        }
        @misc{FragmentRepositoryLSUGVC2017,
          title = {Fragment Image Repository at LSU-GVC},
          author = {Xin Li},          
          howpublished = {http://www.ece.lsu.edu/xinli/FragmentImageRepository/index.html},
          month = {June},
          year = {2017},
        }

Acknowledgements

This work is supported by NSF IIS-1320959.

Other: