Skip to content
Eric Hare

Iowa State University · 2012 — 2017 · Statistics & Computer Science

Matching spent bullets to the guns that fired them — statistically, reproducibly, in code.

My dissertation replaced a firearms examiner's judgement call through a comparison microscope with an open, end-to-end statistical pipeline operating on 3D surface scans. The resulting methods, software, and publications are now part of the CSAFE research program and have been cited in the ongoing debate over the scientific foundation of pattern-matching forensic disciplines.

Read the thesis AOAS paper bulletxtrctr
5
Years
2012 — 2017 at Iowa State.
2
R packages
bulletr and bulletxtrctr.
1
ASA award
Imaging Section Student Paper, 2016.
3D
Microns of striation
Scans resolve sub-micron barrel tool marks.

The problem

Pattern matching without statistics is just pattern matching.

When a bullet is fired, the barrel leaves fine striations on its surface — tool marks that are, in principle, characteristic of that barrel. Traditional firearms identification asks a human examiner to look through a comparison microscope and decide whether two bullets match. The decision is binary, the rationale is verbal, and the error rate has historically been opaque.

The 2009 NAS report and 2016 PCAST report both singled out pattern-matching forensic disciplines for lacking empirical foundations. You cannot argue about error rates on a method until the method is written down precisely enough to run twice. My dissertation set out to write it down.

The pipeline

Scan. Extract. Align. Score.

Every step is code you can run, not a judgement call you can't audit.

  1. Scan

    High-resolution 3D surface scans of spent bullets — sub-micron vertical resolution across the land-engraved area.

  2. Extract

    Isolate land-engraved areas, detrend the bullet curvature, and pull out the 1D striation signature for each land.

  3. Align

    Pairwise cross-correlation aligns striation signatures between lands, robust to shifts and degradation.

  4. Score

    A random-forest model over features of the aligned signatures yields a match probability you can defend in court.

Synthetic striation signature — illustrative amplitude × position along land

Two signatures overlaid. Peaks and valleys align where the same barrel feature scored both bullets — the cross-correlation between them is what the classifier ultimately turns into a match probability.

What shipped

Software and papers, not just a thesis.

bulletr

Reference R package for reading 3D bullet scans, extracting striations, and scoring matches — the first open, reproducible implementation of the pipeline.

bulletxtrctr

Feature-extraction toolkit separated from the core package for maintainability and reuse across CSAFE tooling.

Methodology

Published in the Annals of Applied Statistics and Law, Probability & Risk, with source code released alongside the papers.

Web prototype

A Shiny app that lets examiners upload scans and step through the matching pipeline interactively.

Publications

The primary sources.

All publications

Timeline

How it played out.

  1. 2012

    Started the PhD at Iowa State, joining Heike Hofmann and Alicia Carriquiry.

  2. 2014

    First working prototype of the striation-extraction pipeline in R.

  3. 2015

    CSAFE launched; the bullet-matching work became part of its founding research portfolio.

  4. 2016

    ASA Imaging Section Student Paper Award for the automated matching methodology.

  5. 2017

    Thesis defended; primary paper published in the Annals of Applied Statistics.

  6. Today

    Still an occasional collaborator on CSAFE’s BulletAnalyzr pipeline and the successor R packages.

People

Advisors and co-authors.

Co-advisor

Heike Hofmann

Professor of Statistics at Iowa State and long-time collaborator on the bullet-matching methodology and the downstream R tooling.

Co-advisor

Alicia Carriquiry

Distinguished Professor and founding director of CSAFE — the Center for Statistics and Applications in Forensic Evidence.

Research home

CSAFE

NIST Center of Excellence. The bullet-matching work continues there under the BulletAnalyzr umbrella.

Still interested?

Happy to talk forensic statistics.

If you're working on pattern-matching evidence, reproducible pipelines, or the R tooling around any of it — I still follow this space.

Get in touch CSAFE