Development of Synthetically Generated LEA Signatures to Generalize Probability of False Positive Identification Estimates

• NCJ Number: 240690
• Author(s): Benjamin Bachrach; Pan Gao; Roger Xu; Wei Wang; Ajay Mishra; Kaizhi Tang; Guangfan Zhang
• Date Published: October 2012
• Length: 72 pages
• Grant Number(s): 2009-DN-BX-K236
• Agencies: NIJ-Sponsored
• Publication Type: Report (Study/Research)
• Award: Development of Synthetically Generated LEA Signatures to Generalize Probability of False Positive Identification Estimates

Annotation

This project developed a bullet data synthesis to generate Land Engraved Area (LEA) signatures by determining the characteristics of each brand and the details of each barrel.

Abstract

This project was prompted by the nature of automated ballistics identification (ABI) systems, which require a large amount of data in order to evaluate performance. It is impractical to fire such a large number of bullets for this purpose. In order to address this circumstance, the current project developed a bullet data synthesis methodology for a class of guns based on a set of fired sample bullets. The results from the sample bullets can be used to generalize to a much larger population of firearms. The LEA signature is a one-dimensional signal computed from a digitized three-dimensional bullet surface. The LEA signature is used for bullet matching in ABI systems. The approach used in this project consists of a deterministic component and a random component. The deterministic component includes periodicities in a base curve profile, and the random component is best represented by the fractal model of an irregular curve. The parameters used in the generation of the base and fractal curves are obtained from signal analysis based on wavelet transformation. The details of this method are described in this report. An optimization procedure was applied to the synthesis process in order to ensure the matched and non-matched correlation distribution of the new synthesized dataset resemble that of the existing data of the same brand. Researchers also extended the data synthesis process from a one-dimensional LEA signature signal to a two-dimensional LEA image and further to a three-dimensional bullet surface. According to preliminary data synthesis results, newly generated LEAs for a brand show strong similarity to the LEAs in the same brand, but no similarity to different brands. 53 figures, 8 tables, and 20 references