Date of Graduation

2015

Document Type

Thesis

Degree Type

MS

College

Eberly College of Arts and Sciences

Department

Mathematics

Committee Chair

Jacqueline Speir

Committee Co-Chair

Gerald Lang

Committee Member

Nicholas Petraco

Abstract

Footwear evidence has tremendous forensic value; it can focus a criminal investigation, link suspects to scenes, help reconstruct a series of events, or otherwise provide information vital to the successful resolution of a case. When considering the specific utility of a linkage, the strength of the connection between the source footwear and an impression left at the scene of a crime varies with the known rarity of the shoeprint itself, which is a function of the class characteristics, as well as the complexity, clarity, and quality of randomly acquired characteristics (RACs) available for analysis. To help elucidate the discrimination potential of footwear as a source of forensic evidence, the aim of this research was three-fold.;The first (and most time consuming obstacle) of this study was data acquisition. In order to efficiently process footwear exemplar inputs and extract meaningful data, including information about randomly acquired characteristics, a semi-automated image processing chain was developed. To date, 1,000 shoes have been fully processed, yielding a total of 57,426 RACs characterized in terms of position (theta, r, rnorm), shape (circle, line/curve, triangle, irregular) and complex perimeter (e.g., Fourier descriptor). A plot of each feature versus position allowed for the creation of a heat map detailing coincidental RAC co-occurrence in position and shape. Results indicate that random chance association is as high as 1:756 for lines/curves and as low as 1:9,571 for triangular-shaped features. However, when a detailed analysis of the RAC's geometry is evaluated, each feature is distinguishable.;The second goal of this project was to ascertain the baseline performance of an automated footwear classification algorithm. A brief literature review reveals more than a dozen different approaches to automated shoeprint classification over the last decade. Unfortunately, despite the multitude of options and reports on algorithm inter-comparisons, few studies have assessed accuracy for crime-scene-like prints. To remedy this deficit, this research quantitatively assessed the baseline performance of a single metric, known as Phase Only Correlation (POC), on both high quality and crime-scene-like prints. The objective was to determine the baseline performance for high quality exemplars with high signal-to-noise ratios, and then determine the degree to which this performance declined as a function of variations in mixed media (blood and dust), transfer mechanisms (gel lifters), enhancement techniques (digital and chemical) and substrates (ceramic tiles, vinyl tiles, and paper). The results indicate probabilities greater than 0.850 (and as high as 0.989) that known matches will exhibit stochastic dominance, and probabilities of 0.99 with high quality exemplars (Handiprints or outsole edge images).;The third and final aim of this research was to mathematically evaluate the frequency and similarity of RACs in high quality exemplars versus crime-scene-like impressions as a function of RAC shape, perimeter, and area. This was accomplished using wet-residue impressions (created in the laboratory, but generated in a manner intended to replicate crime-scene-like prints). These impressions were processed in the same manner as their high quality exemplar mates, allowing for the determination of RAC loss and correlation of the entire RAC map between crime scene and high quality images. Results show that the unpredictable nature of crime scene print deposition causes RAC loss that varies from 33-100% with an average loss of 85%, and that up to 10% of the crime scene impressions fully lacked any identifiable RACs. Despite the loss of features present in the crime-scene-like impressions, there was a 0.74 probability that the actual shoe's high quality RAC map would rank higher in an ordered list than a known non-match map when queried with the crime-scene-like print. Moreover, this was true despite the fact that 64% of the crime-scene-like impressions exhibit 10 or fewer RACs.

Download

Share

COinS