This project looks at the challenges involved in the automatic reconstruction of strip (vertically cut) and cross (both vertically and horizontally cut) shredded documents. The unshredding problem is of interest in the fields of forensics, investigative sciences, and archaeology.
All stages of the unshredding pipeline are analysed, starting from scanned images of shreds and ending with reconstructed documents. The current bottlenecks in this pipeline are identified and solutions are proposed.
The original contributions of this project include a probabilistic scoring function which outperforms the standard cost functions used in literature, a refinement upon a previously proposed, graph-inspired, search heuristic and a tractable up/down orientation method for strip-cut shreds.
- UnshredderThesis.pdf for the full report on this project (68 pages). Available at: http://www.cl.cam.ac.uk/~rr463/Shredder_Thesis.pdf
- The paper: A Composable Strategy for Shredded Document Reconstruction, R Ranca, I Murray Computer Analysis of Images and Patterns, 324-331 (8 pages). Available at: http://www.cl.cam.ac.uk/~rr463/Shredder_CAIP.pdf
- The workshop paper: A Modular Framework for the Automatic Reconstruction of Shredded Documents, R Ranca Workshops at the Twenty-Seventh AAAI Conference on Artificial Intelligence (2.5 pages). Available at: http://www.cl.cam.ac.uk/~rr463/Shredder_AAAI.pdf