Citation

Chuan ZL, Ghani NAM, Liong CY, Jemain AA. Sains Malays. 2013; 42(9): 1339-1344.

Copyright

(Copyright © 2013)

DOI

unavailable

PMID

unavailable

Abstract

Since the number of crimes involving firearms is becoming rampant, identification of firearms used by criminals is a crucial step in the court. Several automatic firearm identification systems have been developed to improve on the traditional investigation method which relies heavily on the expertise of the forensic ballistics experts. An important step in automatic firearm identification is partitioning of the region of interest (ROI) based on the position of the anchor point (PAP) within the circular boundary of a firing pin impression. However, in the previous studies, the methods used to determine the PAP of a circular boundary are very complex and time consuming. This study explored algorithms that are efficient and able to detect the anchor point of a circular boundary automatically. The proposed algorithms are a combination of sharpening spatial filter, histogram normalization, thresholding and an unweighted least square estimator. Two popular threshold selection methods, namely clustering-based and entropy-based threshold selection methods, have been investigated and compared. In addition, exploration on the effects of size and shape of ROI on the firearm classification accuracy rates were discussed. A total of 747 images of circular boundary firing pin impression produced by five different pistols of the same model were used to test the proposed algorithms. Encouraging classification rates of the firing pin impression images (> 95%) were achieved with the proposed algorithms. This study also found that the size and the shape of the ROI partition have a direct effect on the firearms classification rates.

Language: ml