论文标题
使用记录链接框架自动匹配墨盒外壳的地形测量
Automatically matching topographical measurements of cartridge cases using a record linkage framework
论文作者
论文摘要
发射枪支在墨盒箱上留下了痕迹，这些盒子据称是独特地识别枪支的。枪支检查员通常使用视觉检查来评估是否从同一枪发射了两个弹药箱，这是一个受到审查的主观过程。可以使用自动化算法以更可重现的方式进行匹配。在本文中，我们开发了比较墨盒案例的地形测量的方法。我们在此上下文中演示了记录链接框架的使用。我们将使用地形测量的性能与较旧的反射率显微镜图像进行比较，并研究了前者产生更准确的比较程度。使用超过1,100个墨盒盒的各种图像集合，我们发现使用地形数据通常可以改善整体性能。数据的某些子集从精确和回忆方面获得了几乎完美的预测性能，而有些则产生极低的性能。需要进行进一步的工作来评估考官是否在某些枪支和弹药组合上面临类似的困难。对于自动方法，需要对其公平和鲁棒性进行更全面的调查，然后才能在实践中部署它们。
Firing a gun leaves marks on cartridge cases which purportedly uniquely identify the gun. Firearms examiners typically use a visual examination to evaluate if two cartridge cases were fired from the same gun, and this is a subjective process that has come under scrutiny. Matching can be done in a more reproducible manner using automated algorithms. In this paper, we develop methodology to compare topographical measurements of cartridge cases. We demonstrate the use of a record linkage framework in this context. We compare performance using topographical measurements to older reflectance microscopy images, investigating the extent to which the former produce more accurate comparisons. Using a diverse collection of images of over 1,100 cartridge cases, we find that overall performance is generally improved using topographical data. Some subsets of the data achieve almost perfect predictive performance in terms of precision and recall, while some produce extremely poor performance. Further work needs to be done to assess if examiners face similar difficulties on certain gun and ammunition combinations. For automatic methods, a fuller investigation into their fairness and robustness is necessary before they can be deployed in practice.
