Classification of pistol via numerical based features of firing pin impression image

摘要

A lot of current crime cases have been reported to involve pistols, among other firearms. The whole firing pin impression image on a cartridge case is one of the most substantial clues for firearms identification. In this study, a total of 16 features of geometric moments up to the sixth order were extracted from the entire firing pin impression images. All five pistols of the Parabellum Vector SPI 9mm model, manufactured in South Africa were used. The pistols were marked Pistol A, Pistol B, Pistol C, Pistol D, and Pistol E. A total of 747 bullets have been launched from the five pistols. Under an initial analysis, Pearson correlation coefficients between all pairs of features have demonstrated that the features were significant and that the features were inter-related. These problematic featureswere solved by dividing the features into subgroups of variables based on the same characteristics under the principle component analysis. The features that are highly correlated were brought together into meaningful components or factors. Discriminant analysis was applied for the identification of the types of pistols used based on the factors obtained. Classification results using cross-validation under the discriminant analysis pointed that 65.7% of the images were rightly classified according to the pistols used.