Automatic Firing Pin Impression Identification based on Feature Fusion of Fractal Dimension and Geometric Moment

Abstract

Automatic firearms identification based on the physical evidence of firing pin impression is very vital for forensic investigation. Currently, due to complex topography of firing pin impression, the firearms identification has been performed manually and the precision of comparisons relies on the human expertise. This approach normally requires a long time to observe through a large number of image database. To overcome this problem, an automatic ballistics identification system using the feature fusion of fractal dimension and geometric moment is proposed. In this study, eight fractal dimension features and 11 geometric moment features were extracted from firing pin impression images of five pistols of the Parabellum Vector SPI 9 mm model. These features were passed to five different machine learning methods for classification. The experimental results indicated that the neural network classifier achieved the highest classification performance of 99.3%, which is a very promising result. In conclusion, the features fusion of fractal dimension techniques and geometrical moments, with neural network as classifier yields impressive results towards automatic pistol detection.