Automated encoding of footwear patterns for fast indexing

The rapid and robust identification of a suspect’s footwear while he/she is in police custody is an essential component in any system that makes full use of the footwear marks recovered from crime scenes. Footwear is an important source of forensic intelligence and, sometimes, evidence. Here, we present an automated system for shoe model identification from outsole impressions taken directly from suspect’s shoes that can provide information in a timely manner, while a suspect is in custody. Currently the process of identifying the shoe model from the 1000 s of recorded model types is a time-consuming manual task. The underlying methodology is based on large numbers of localized features located using MSER feature detectors. These features are transformed into robust SIFT descriptors and encoded relative to a feature codebook forming histogram representations of each shoe pattern. This representationist facilitates fast indexing of footwear patterns whilst a finer search proceeds by comparing the correspondence between footwear patterns in a short-list through the application of modified constrained spectral correspondence methods. The effectiveness of this approach is illustrated for a reference dataset of 374 different shoe model patterns, from which 87% first-rank performance and 92% top-eight rank performance are achieved. Practical aspects of the system and future developments are also discussed.