Cross-Domain Forensic Shoeprint Matching
icon We investigate the problem of automatically determining what type of shoe left an impression found at a crime scene. This recognition problem is made difficult by the variability in types of crime scene evidence (ranging from traces of dust or oil on hard surfaces to impressions made in soil) and the lack of comprehensive databases of shoe outsole tread patterns. We find that mid-level features extracted by pre-trained convolutional neural nets are surprisingly effective descriptors for these specialized domains. However, the choice of similarity measure for matching exemplars to a query image is essential to good performance. For matching multi-channel deep features, we propose the use of \emph{multi-channel normalized cross-correlation} and analyze its effectiveness. Finally, we introduce a discriminatively trained variant and fine-tune our system end-to-end, obtaining state-of-the-art performance.

Download: pdf

Text Reference

Bailey Kong, James Steven Supan\vc i\vc , Deva Ramanan, and Charless C. Fowlkes. Cross-domain forensic shoeprint matching. In British Machine Vision Conference (BMVC). 2017.

BibTeX Reference

@inproceedings{KongSRF_BMVC_2017,
    author = "Kong, Bailey and Supan{\vc}i{\vc}, James Steven and Ramanan, Deva and Fowlkes, Charless C.",
    title = "Cross-Domain Forensic Shoeprint Matching",
    booktitle = "British Machine Vision Conference (BMVC)",
    year = "2017"
}