U.S. flag

An official website of the United States government, Department of Justice.

NCJRS Virtual Library

The Virtual Library houses over 235,000 criminal justice resources, including all known OJP works.
Click here to search the NCJRS Virtual Library

Quantitative Measures in Support of Latent Print Comparison

NCJ Number
241288
Author(s)
Sargur N. Srihari
Date Published
February 2013
Length
63 pages
Annotation
This research addresses the evaluation of three quantitative measures: rarity of features, confidence of opinion and a probabilistic measure of similarity.
Abstract
Latent prints of friction ridge impressions have long been useful in identification, and the methodology of examining latent prints, known as ACE-V (analysis, comparison, evaluation, and verification), has been well-documented. This research addresses how to model probability distributions of features, how they can be used to determine the rarity of evidence (as measured by the probability of random correspondence in a database of given size), how to make such evaluations computationally tractable, how rarity can be combined with similarity (between the evidence and a known) to determine the confidence of a conclusion, and how to obtain a probabilistic measure of similarity. The need to quantify confidences within ACE-V has been articulated in several recent influential reports to strengthen the science of friction ridge analysis. Rarity is difficult to compute due to the large number of variables and high data requirements. The proposed solution uses probabilistic graphical models to represent spatial distributions of fingerprints represented at level 2 details (minutiae). First, the minutia coordinate system is transformed into standard position based on a point of high curvature, viz., core point; statistical regression (based on a Gaussian process formulation and a training set of latent prints) is used to estimate the core point. A directed probabilistic graphical model is constructed using inter-minutia dependencies and minutia confidences. The resulting model is used to determine the probability of random correspondence of the evidence in a database of n prints. The developed methods can be of potential use in examiner training, presentation of opinion and validating examination procedures. Tables, figures, and references