Firearms Identification Using Pattern Analysis and Computational Modeling.

This paper describes a computational approach to the analysis of firearms forensic images, in particular, bullet images. Expended bullets usually bear characteristic surface markings associated with the mechanisms of the firearm that fired them. The traditional method of comparing bullet and cartridge markings involves the labor-intensive process of visual identification through a comparison microscope. This project proposes that the same process can be duplicated, and accelerated, using a low-cost video imaging and image processing system. The development of this project requires three important phases - collection, processing, and analysis. In a laboratory scenario, the barcodes of the crime bullet and test bullet images are matched against each other. The barcode of the crime bullet is also matched against the barcodes of other potentially similar bullet images from the image database. A probability of match is generated for each comparison, and bullet images with a low probability of match are ignored. This effectively reduces the firearms examiners workload. The bullet images with a high probability of match are retrieved, and displayed simultaneously to confirm the match.