AbstractThis thesis tackles the challenges presented by shoemark evidence assessment within forensic science as summarised in the recent (2010) "RvT" judgement Modern evidence types, such as DNA, have raised expectations around the use of transparent and objective scientific frameworks within forensic evidence assessment.
Initially a new dataset of temporal shoemark images is presented for use in shoe wear experiments, captured over a period of five months. Using this data two computational solutions are designed; first in the spatial domain using SIFT, an approach previously proven in shoemark classification, and subsequently using a novel ridge detection algorithm within the frequency domain. Finally, a highlevel, conceptual framework is presented for integrating computational techniques with current academic and industry research in the area of forensic
shoemark evidence assessment.
Experiments with the SIFT-based algorithm showed that SIFT feature matching, although effective in shoe pattern classification, is not adept at detecting corresponding wear features and the results were disappointing. However, the frequency domain ridge detection algorithm produced more promising results, which were enhanced further with an additional image fusion stage applied when multiple source images were available.
As they stand the results show that the algorithms do not yet enable wear estimates to be made reliably. Nevertheless, when coupled with the generalised, and novel, wear model, the methodology provides an extensible platform for further work in the field of computational forensics with respect to shoe sole wear.
|Date of Award||Jul 2014|
|Supervisor||Patrick Crookes (Supervisor)|