Shoeprints are routinely left at crime scenes and are reported to be present more frequently than fingerprints. It has been reported that 35 percent of crime scenes present shoe marks that can be recovered and used as forensic evidence. During investigations, a scene of crime shoeprint can be matched against a database of known shoeprints in order to identify the brand and the model of the corresponding shoe. TIlls is known as shoeprint classification and is, currently, performed manually or using some semi-automatic systems. These current approaches are time consuming and are not very reliable. Thus, the development of automatic shoeprint classification methods would offer valuable assistance to forensic scientists. TIlls thesis addresses the task of automatic shoeprint classification and its related challenges. This includes the problem of classifying partial, noisy and/or blurred shoeprint images. The issues of invariance to geometric distortions, e.g. translations and rotations, as well as rapid classification are also considered. The thesis proposes a number of different ideas and methods for the automatic classification of distorted shoeprint images including the use of Fourier-Mellin transform, modified phase-only correlation and two-dimensional advanced correlation filters. It also investigates the use of multiple one-dimensional correlation filters and classifier combination techniques, such as algebraic rules, Decision Templates and Support Vector Machine based combiners. The experimental results suggest that the investigated correlation-based methods can offer high accuracies when classifying low quality shoeprint images while providing tolerance to geometric distortions.
| Identifer | oai:union.ndltd.org:bl.uk/oai:ethos.bl.uk:557608 |
| Date | January 2012 |
| Creators | Gueham, M. |
| Publisher | Queen's University Belfast |
| Source Sets | Ethos UK |
| Detected Language | English |
| Type | Electronic Thesis or Dissertation |
Page generated in 0.1181 seconds