Application of Spatial Statistics to Latent Print Identifications
Towards Improved Forensic Science Methodologies
Stephen J. Taylor, Emma K. Dutton, Patrick R. Aldrich, Bryan E. Dutton
See also the instructional videos on "Fingerprint Comparison"
Abstract
In 2010 we initiated a research project to address criticisms raised in a 2009 National Academy of Sciences (NAS) report regarding the presumption of fingerprint uniqueness and the reliability of latent print identifications using the ACE-V methodology (National Research Council 2009). This project addresses the question of fingerprint uniqueness (i.e., the discriminating value of the various fingerprint ridgeline features) by statistically evaluating the spatial distribution of these features. The purpose of the project was to review the latent print ACE-V comparison methodology to ascertain the fingerprint features considered during the comparison process and apply principles of spatial analyses to calculate false-match probabilities. The objectives were to spatially analyze fingerprint features (e.g., minutiae and ridge lines) using Geographic Information Systems (GIS) techniques and empirically derive probabilities to provide a quantitative measure of the discriminating value of the various ridgeline features. The resultant probabilities are applicable for subsequent qualification of latent print comparison conclusions.
Project methods included spatial pattern characterization using GIS, geometric morphometric (GM) analysis, and the calculation of false-match probabilities using Monte Carlo (MC) simulations. A data set of digitized fingerprints from the Oregon population was compiled and spatially analyzed utilizing GIS software to place minutiae and ridge line features in a common Cartesian coordinate system. The parameters of these fingerprint features, including minutiae location, direction and minutiae ridgeline configurations, were evaluated. Geometric morphometrics was used to study shape variation between and among fingerprint pattern types. GIS-based procedures were established for the selection of landmarks and semi-landmarks, the superimposition of fingerprint images, the visualization of shape change, the ordination of superimposition data, and the application of multivariate statistics. Using MC simulations, random-match probabilities were calculated to evaluate the spatial configurations of minutiae within and between pattern types to quantitatively evaluate the discriminating value of fingerprints features; that is, do two fingerprints or two regions of different fingerprints have the same spatial distribution of minutiae and ridgelines? MC simulations were performed using 3, 5, 7 and 9 minutiae with other minutiae attributes chosen for additional match criteria.
GIS results showed there was a greater density of minutiae and ridgelines below the core compared to above the core, regardless of pattern type. However, the distributions of bifurcations and ridge endings were more similar within any pattern type rather than among them. Also, pattern types with comparable ridge flow (e.g., right and left slant loops, and whorls and double loop whorls) had greater similarity between them when comparing various metrics such as axis dimensions and Thiessen polygon ratios. GM results demonstrated little shape variation among fingerprints of the same pattern type with the greatest shape variation associated with the deltas. Additional GM spatial analyses suggested a very high degree of shape consistency between left and right slant loops and between whorls and double loop whorls. MC simulations showed that the probability of random minutiae correspondence drastically decreased as the fingerprint attribute criteria (e.g., minutiae type and direction) increased. In addition, increasing the number of minutiae and fingerprint attributes applied in searches away from the core and delta regions yielded lower probabilities for a false match. However, results demonstrated that minutiae spatial distributions in regions around and below the core were not always unique.
Fingerprint characterization of ridgeline minutiae configurations and establishing random-match probabilities when using specified features quantitatively describe the discriminating value of these fingerprint ridgeline features. As such, random-match probabilities will allow the latent print examiner to qualify their comparison conclusions.