Classification of firing pin impressions using HOG-SVM

Area of Science:

• Forensic Science
• Computer Vision
• Machine Learning

Background:

• Firearm examination is crucial for criminal investigations, particularly in linking cartridge cases to suspect firearms.
• Firing pin impressions on cartridge cases are key identifying marks used in forensic analysis.
• Traditional methods rely on manual comparison, which can be time-consuming and subjective.

Purpose of the Study:

• To develop and evaluate a computational method for classifying firing pin impressions.
• To assess the accuracy of a computer vision algorithm (Histogram of Oriented Gradients - HOG) combined with a machine learning method (Support Vector Machines - SVMs) for this task.
• To compare the performance of the HOG-SVM method against other feature extraction algorithms.

Main Methods:

• Nine Ruger model 10/22 semiautomatic rifles were used, firing 50 cartridges each.
• Cartridge cases were collected, and firing pin impressions were cast and photographed.
• Images of firing pin impressions were analyzed using the Histogram of Oriented Gradients (HOG) algorithm and Support Vector Machines (SVMs).

Main Results:

• The developed HOG-SVM method achieved a classification accuracy of 93% for firing pin impressions.
• The HOG-SVM method demonstrated superior performance compared to other feature extraction algorithms tested.
• The study successfully demonstrated the potential of computational methods in firearm examination.

Conclusions:

• The HOG-SVM approach offers a reliable and accurate computational tool for classifying firing pin impressions.
• This method can significantly assist firearm examiners in associating cartridge cases with specific firearms at crime scenes.
• Automated analysis of ballistic evidence holds promise for enhancing the efficiency and objectivity of forensic investigations.