This webinar is on-demand and available immediately.

Automatic cartridge case comparison algorithms can be used to obtain an objective measure of similarity for two fired cartridge cases. For example, an algorithm might compare 3D topographic scans of two cartridge case surfaces and return an estimated probability that the cartridge cases originated from the same firearm. However, the manner in which a comparison algorithm arrives at such a similarity score can be difficult to interpret or explain, even for the original authors of the algorithm.

We introduce a set of visual and interactive diagnostic tools that are useful for comparing cartridge case breech face impressions. These tools use intuitive visualizations to uncover similarities and differences in the impressions on two cartridge case surfaces. We demonstrate how the diagnostic tools are useful in explaining the behavior of comparison algorithms. Specifically, we apply the visual diagnostics to each stage of the Automatic Cartridge Evidence Scoring (ACES) algorithm that pre-processes, compares, and scores two cartridge case scans.

We also compute numerical features based on qualitative observations made about visual diagnostics and use these features in the ACES algorithm to estimate common source probabilities for pairs of cartridge cases. Inclusion of these features results in an approximately one-third reduction in the classification error rate of the ACES algorithm.

We implement these visual diagnostics in an interactive web application called cartridgeInvestigatR that allows non-programmers to apply, explore, and diagnose automatic comparison algorithms. We provide these diagnostic tools to the forensics community as free and open-source software to promote explainability and transparency of automatic comparison algorithms.

Presenter: Joseph (Joe) Zemmels

Recorded: 03/09/2023