Automated Bullet Matching

When a bullet is fired from a gun, the barrel leaves fine striations on the bullet’s surface. Traditionally, matching a recovered bullet to a particular gun is done by a firearms examiner looking through a comparison microscope and making a judgement call. My dissertation — and the ongoing CSAFE research program it fed into — builds reproducible, statistical methods for doing that matching from high-resolution 3D surface scans.

The practical artifacts:

• bulletr — the reference R package for reading 3D bullet scans, extracting striations from land-engraved areas, aligning them, and scoring match probabilities. (CRAN-archived in late 2025 because a transitive dependency was archived; actively maintained on GitHub.)
• bulletxtrctr — the feature-extraction toolkit.
• Published methodology in Annals of Applied Statistics (“Automatic Matching of Bullet Land Impressions”) and Law, Probability & Risk (“Algorithmic Approaches to Match Degraded Land Impressions”).
• Webapp prototype at labs.omnianalytics.org/bullet-analyzer.

The work won the ASA Imaging Section Student Paper Award (2016) and is now part of the CSAFE BulletAnalyzr pipeline I still push to occasionally.

Why it matters

Firearms examination is one of the pattern-matching forensic disciplines that has been repeatedly flagged for lacking statistical foundations (NAS 2009; PCAST 2016). A reproducible, open-source, quantitative pipeline isn’t a complete answer to those critiques, but it’s a necessary piece — you can’t argue about error rates on a method until the method is written down precisely enough to run twice.