Author ORCID Identifier

https://orcid.org/0009-0000-3386-6423

Semester

Summer

Date of Graduation

2026

Document Type

Thesis

Degree Type

MS

College

Eberly College of Arts and Sciences

Department

Forensic and Investigative Science

Committee Chair

Keith Morris

Committee Co-Chair

Tina Moroose

Committee Member

Theunis Brits

Abstract

Cartridge cases recovered from discharged firearms contain tool marks produced by the internal mechanisms of said firearm. Firearm and tool mark examiners com pare these marks to determine whether two cartridge cases originated from the same firearm. Traditional comparison methods have been criticized for their subjective nature and reliance on examiner interpretation rather than objective decision thresholds. The Congruent Matching Cells (CMC) algorithm was developed as a quantitative, objective method for tool mark comparison. CMC is a surface topography correlation algorithm that partitions a tool mark into small regions, or cells, and evaluates their similarity using the areal cross-correlation function (ACCF). The number of congruent matching cells is then used as a similarity score to assess whether two tool marks share a common source. Previous studies have demonstrated the effectiveness of CMC for centerfire cartridge cases and fired projectiles. This research had two objectives: (1) to optimize CMC parameters for rim fire firing pin impressions and (2) to optimize the Gaussian preprocessing filters used for centerfire breechface impressions. Optimization was performed using the Nelder–Mead method. For rimfire impressions, the optimized parameters were cell size and the threshold for inclusion, while the effects of objective magnification were also evaluated. For centerfire breechface impressions, the optimized parameters were the Gaussian low- and high-pass filters. The first hypothesis proposed that the CMC algorithm could reliably distinguish known matching from known non-matching rimfire firing pin impressions. How ever, the irregular surface topography of the impressions resulted in unfavorable performance, providing evidence against this hypothesis. The second hypothesis proposed that improved Gaussian filter parameters could be identified. Although three optimal parameter sets outperformed the NIST-recommended parameters on the training set, the NIST parameters produced superior performance on the test set, providing evidence both supporting and refuting the hypothesis. Ultimately, given the training set was comprised of more comparisons, it is determined that the hypothesis is supported, as larger sample sets are more robust

Download

Share

COinS