Supervisor: Manuel Keglevic
One of the most frequently secured types of evidence at crime scenes are footware impressions. Identifying the brand and model of the footware can be crucial to narrowing the search for suspects. This is done by forensic experts by comparing the evidence found at the crime scene with a huge list of reference impressions. In order to support the forensic experts an automatic retrieval of the most likely matches is desired.
The goal of this work is to apply pattern matching of such images. As a starting point, the algorithm proposed by Kortlylewski et al. [1,2] should be implemented and evaluated. Further, it should be studied if the results could be improved using approaches based on machine learning (deep learning).
- Literature Review – getting to know the algorithms
- Data Preparation
- Written Report/Thesis and final presentation
- Experience with Matlab or Matlab-like environments (NumPy, torch7, …)
- Machine Learning knowledge beneficial
 Adam Kortylewski, Thomas Albrecht, Thomas Vetter, “Unsupervised Footwear Impression Analysis and Retrieval from Crime Scene Data“, Asian Conference on Computer Vision (ACCV), Workshop on Robust Local Descriptors, 2014
 Adam Kortylewski, Thomas Vetter, “Probabilistic Compositional Active Basis Models for Robust Pattern Recognition“, BMVC, 2016