Degree Type

Dissertation

Date of Award

2017

Degree Name

Doctor of Philosophy

Department

Statistics

Major

Statistics

First Advisor

Max D. Morris

Abstract

If a striated tool such as a screw driver is used to commit a crime, often there is a tool mark left behind as evidence. It is then the job of a forensic tool mark examiner to compare this crime scene tool mark to tool marks made by a suspect's tool to determine if they match, i.e were made by the same tool. Forensic tool mark examiners say that these striated tool marks are comprised of 'class' and 'individual' characteristics where class characteristics are traits common to a large number of tools such as the width of a screwdriver head and individual characteristics are traits unique to a specific tool such as imperfections and wear patterns in the surface of a screwdriver. Examiners first compare marks according to their class characteristics and if they match, they continue to compare the individual characteristics. If the class characteristics do not match, it is concluded the marks were not made by the same tool. Many of the algorithms being developed to remove the subjective nature of an examiner's comparison ignore the distinction between class and individual characteristics and attempt to directly compare the marks visually, or by applying some quantitative similarity index. We have developed a procedure for comparing tool marks that initially decomposes a digitized tool mark into class and individual components and applies a fixed width window correlation separately to each component. Based on the offsets (or registration) producing the maximized correlation (optimal offsets) and the correlation at the remaining offsets, we formulate hypothesis tests (Different Tool vs. Common Tool) with test statistics and p-values based on the distance between the optimal offsets for the two components, or by setting a threshold for correlations between the individual component series. Additionally, we have developed a simulation based approach for a test based on the maximized correlation between two tool marks. Finally, we review the method of Chumbley et al. (2010) and propose possible improvements.

Copyright Owner

Jeremy R Hadler

Language

en

File Format

application/pdf

File Size

100 pages

Share

COinS