Background
One of OAITI’s earliest strategy partners is CSAFE. CSAFE was created with a mission of strengthening forensic science using statistical foundations and open algorithms. OAITI has partnered with CSAFE on bullet matching routines, which are designed to perform a statistical assessment of the likelihood that two bullets were fired from the same gun barrel. We partner with CSAFE on these algorithms in order to ensure that valid statistical procedures are followed, eliminating the risk of wrongful convictions and lessening the cost and time burden on the judicial system.
Data
OAITI utilizes the National Ballistics Toolmark Research Database. This database is a joint effort between agencies at NIST, CSAFE, and other organizations to collect structured, open forensics science data which can be used to train and refine algorithms that enhance the forensics community. Our primary focus is on the 3D bullet scan data contained within this database. We augment this data with other scans produced and collected using advanced imaging technology at CSAFE.
Methodology
Our bullet matching procedure uses the following steps:
1. Identify a cross-section of the bullet land to extract a profile
2. Perform statistical procedures to smooth out the profile and focus on the striations
3. Identify the locations and properties of every striation
4. Use cross-correlation to pairwise-align bullet profiles
5. Compute features assessing the degree of agreement between the profiles
6. Use a Random Forest to predict the probability of a match
Outcome
The bullet matching algorithms achieve state-of-the-art accuracy on a small test database of bullets. They are able to correctly classify every pairwise bullet as match / non-match, and over 99% of every pairwise bullet-land, without any human intervention.
Social Benefit
Several studies have been released suggesting that traditional forensic science procedures are lacking in scientific rigor (e.g., NAS 2009) We, along with CSAFE, share the goal of enhancing these procedures to eliminate the risk of wrongful convictions, and to streamline the process of criminal justice proceedings where possible. Though the input of experts in the forensics science community will likely always be an integral part of the criminal justice system, a focus on objective, statistical procedures can bring much needed transparency and openness to a process that can mean life or death for those being tried.
Links
Automatic Matching of Bullet Land Impressions, Hare, Hofmann, Carriquiry (2017)