Shoe prints gathered from crime scenes are considered important forensic evidence in legal courts. To date, forensic examiners have to manually compare shoe prints with a large set of reference shoe prints. Since this is a time consuming task an assisiting pattern recognition system is required.

The goal of this project is to design a pattern recognition system to identify a suspect's shoe by analysing local features of the corresponding shoe print pattern. In a first step, the shoe print will be digitalized by either a camera or a scanner. Subsequently, image features are extracted by analysing locally distinctive points such as corners or homogeneous regions and their spatial relationships. Several feature extraction methods can be applied for that task, which will be evaluated extensively during the project. Furthermore, a representation has to be defined which can encode various spatial relationships such as distance, relative dimensions and positions.
Based on the representation and the shoe print image, a particular shoe model will be identified by a suitable classifier. The pattern matching system has to be capable of solving the desired tasks on a datasbase with several thousand reference shoeprints with a high accuracy and in a feasible time frame.