Analysis of Geometric Moments as Features for Identification of Forensic Ballistics Specimen.
ABSTRACT Firearm identification is one of the most essential, intricate and demanding tasks in crime investigation. Every firearm,
regardless of its size, make and model, has its own unique ‘fingerprint’ with respect to the marks on fired bullet and cartridge
cases. In this study, we investigate the features extracted from the images of the centre of the cartridge case in which firing
pin impression is located. Geometric moments up to the sixth order were computed to obtain the features based on a total of
747 cartridges case images from five different pistols of the same model. These sixteen features were found to be significantly
different using the MANOVA test. Correlation analysis was used to reduce the dimensionality of the features into only six
features. Classification results using cross-validation show that about 74.0% of the images were correctly classified and
this demonstrates the potential of using moment based features for firearm identification.
- SourceAvailable from: sciencedirect.com[show abstract] [hide abstract]
ABSTRACT: Many heavy crimes committed such as murders or robberies frequently involve firearms, particularly pistols. In order to solve the crime cases, firearm identification is becoming vital. Unique marks are left on the bullet and the cartridge case when a firearm is fired. The firing pin impression is one of the most vital marks on any cartridge case. In this study, a total of 68 features of firing pin impression images – 20 basic statistical features, and 48 geometric moment features up to the sixth order – were extracted from three regions of the firing pin impression image, namely whole, centre and ring images. Five different types of pistol of the Parabellum Vector SPI 9 mm model were tested, where 50 bullets were fired from each pistol. Preliminary analysis using Pearson correlation shows that the features are significantly highly correlated. Therefore principal component analysis (PCA) was used to analyze the interrelationship among the features and combine them into a smaller set of factors while maintaining maximum information of the original patterns. PCA has reduced the dimensionality of the features into nine significant components of features. Discriminant analysis was used to identify the types of pistols used based on the new components. A total of 85.2% of the images were correctly classified according to the pistols used using cross-validation under discriminant analysis. The result demonstrates the potential of using PCA to reduce the dimensions of the numerical features towards an efficient firearm identification system.Procedia Computer Science. 13:144–151.