B0353
Title: FICSing forensic footwear comparison
Authors: Steven Lund - National Institute of Standards and Technology (United States) [presenting]
Adam Pintar - National Institute of Standards and Technology (United States)
Abstract: Footwear impressions are among the most commonly available types of forensic evidence, estimated to be recoverable from roughly 40\% crime scenes. However, progress in footwear impression comparison algorithms has lagged behind that of other pattern comparison disciplines such as fingerprints, handwriting, firearms, and face recognition. This is due to the wide variety of signal, noise, and structured backgrounds encountered in footwear impression images combined with limited available databases. Computational challenges and solutions involved in developing the NIST footwear impression comparison system (FICS) are overviewed. It illustrates how discrimination performance can improve using image segmentation driven by unsupervised (e.g., superpixel) and supervised (e.g., U-Net) learning approaches. Its discrimination performance is additionally compared to that of professional examiners.
