Generalizable features-based method for breech face comparisons

Breech face impressions are marks left by the impact of a bullet casing against the breech face due to the reactive force of the gunpowder explosion, which is essential evidence linking firearms to fired bullet casings. Aiming at the problem that traditional methods exhibit a high error rate and cannot be applied to both granular and striated impressions, a learnable image matcher designed on the core principle of generalization (OmniGlue) is introduced for the automatic comparison of breech face impressions. The method employs an attention mechanism to propagate information across the built graphs, removes irrelevant key points and separates the local feature descriptors from the position information of key points during information propagation, thereby reducing the dependence of the feature descriptors on position information. Position information is not merged into the local descriptors used for matching, enhancing the generalization of these descriptors and improving the model’s performance across different domains. The method is validated using the Fadul and Hamby datasets provided by the National Institute of Standards and Technology (NIST). The results show that the method is effective for both granular and striated marks, exhibiting a lower identification error rate and superior performance compared to traditional methods.