Towards automated forensic fracture matching of snap-off blade knives

An interesting problem that has concerned forensic scientist for many years, is their need for accurate, reliable and objective methods for performing fracture matching examinations. The aim of these fracture matching methods is to determine if two broken object halves can be matched together, e.g., when one half is recovered at a crime scene, while the other half is found in the possession of a suspect. In this paper we discuss the use of a commercial white-light profilometer system for obtaining 2D/3D image surface scans of multiple fractured objects. More specifically, we explain the use of this system for digitizing the fracture surface of multiple facing halves of several snap-off blade knives. Next, we discuss the realization and evaluation of several image processing methods for trying to match the obtained image scans corresponding to each of the broken off blade elements used in our experiments. The algorithms that were tested and evaluated include: global template matching based on image correlation and multiple template matching based on local image correlation, using so-called "vote-map" computation. Although many avenues for further research still remain possible, we show that the second method yields very good results for allowing automated searching and matching of the imaged fracture surfaces for each of the examined blade elements.