As one form of valuable forensic evidence, the shoeprint marks left at crime scenes have often been ignored. This has now changed as footwear marks are more likely to be recovered from some crime types than fingerprints. Compared to conventional biometric evidence such as DNA traces and fingerprints, it is harder for suspects to conceal their shoeprints. To improve the utilization of footwear impressions in forensic investigations and boost the analysis speed, shoeprint recognition systems have been recently reported. However, without the capability of tackling shoeprints left by the same shoes in different worn conditions, current shoeprint recognition systems may suffer considerable performance degradations in real-world application. We believe that a shoeprint recognition system capable of using 3D depth information from footwear outsoles to produce 2D shoeprints with expected worn specifications is a practical solution to these limitations. In this thesis, an enhanced hybrid method is proposed for 3D outsole feature classification and extraction. By carrying out analysis in terms of curvature and frequency attributes, our proposed method is able to categorize 3D outsole features into two types - Printable and Unprintable 3D Features, according to their contribution in leaving 20 impressions on a surface, and produce initial extraction results. Finally, such rudimentary results are refined by a watershed-based post-processing algorithm to produce elaborate 2D shoeprints. The superior performance of this proposed algorithm is confirmed by both synthetic and real outsole based experiments under various distortions. To exploit the utilization of 20 shoeprints extracted from the 3D outsole models in real world applications, an image patch-based system for shoeprint matching is also proposed. In this method, a mutually collaborative approach is proposed and - 1 - employed for the image patch binarisation operation. With the assistance of the shape context approach and the coherent point drift ((PD) method, skeletons derived from the extracted 2D shoeprint and a grayscale image in the reference database are aligned, and the relevant result is used to evaluate the similarity between these two images. Tested under various distortions, the proposed system demonstrates the feasibility of matching binary shoeprint images to their grayscale counterparts. Finally, this thesis offers some suggestions for future research in this field. ·2·
| Identifer | oai:union.ndltd.org:bl.uk/oai:ethos.bl.uk:575381 |
| Date | January 2012 |
| Creators | Gao, Bo |
| Publisher | University of Sheffield |
| Source Sets | Ethos UK |
| Detected Language | English |
| Type | Electronic Thesis or Dissertation |
Page generated in 0.002 seconds