Quantitative Assessment of the Individuality of Friction Ridge Patterns

• NCJ Number: 227842
• Author(s): Sargur N. Srihari
• Date Published: March 2009
• Length: 94 pages
• Grant Number(s): 2005-DD-BX-K012
• Agencies: NIJ-Sponsored
• Publication Type: Report (Study/Research)
• Award: Quantitative Assessment of the Individuality of Friction Ridge Patterns

Annotation

This project was conducted in order to increase understanding of the discriminative power of friction ridge patterns in fingerprints by using computational approaches.

Abstract

A study of the fingerprints of twins using automatic fingerprint matching algorithms found that although friction ridge patterns of twins are more similar than in the general population, they are still distinctive to each twin. This finding strengthens the argument that fingerprints can accurately identify individuals. In modeling the probability distribution of fingerprints from which the probability of random correspondence of fingerprints is determined (the "generative" approach to individuality); the study found that the individuality of a forensic modality could be established by computing the probability of random correspondence of two pieces of evidence directly from the underlying probability distribution of evidence features. The distribution can be modeled as a mixture distribution whose parameters can be determined from a database of fingerprints. A model that consisted of only minutiae that was then expanded to include ridge information as well determined that the probability of random correspondence with ridges was much lower than with minutiae alone. These probabilities were quantified in terms of the available number of minutiae and ridge points. A third approach used new algorithms for feature extraction and classification in order to determine error probabilities ("discriminative" approach to individuality). This involved two new approaches to automatic fingerprint comparisons: the use of likelihood methods instead of the standard receiver operating characteristics (ROC) methods; and using ridge information in addition to minutiae in fingerprint comparison. Both discriminative studies produced improved performance, particularly when there were fewer minutiae available in the input, as in the case of latent prints. 13 tables, 42 figures, and a 70-item bibliography