Statistical analysis of fingerprint first-level detail using Bayesian networks

Abstract

This study presented a large-scale statistical examination of 168,974 tenprint records to evaluate whether pattern distribution across the fingers is random or exhibits structured interdependence, and whether sex-related differences in pattern frequency exist. Two Bayesian networks were empirically developed and validated to model the relationships between the pattern types of different fingers. The first network focused specifically on the occurrence of whorls and was evaluated relative to established frequencies of the Henry primary classification system, revealing expected relationships between pattern types, but also extending beyond, traditional classification approaches. The second network incorporated all major fingerprint pattern types to model probabilistic dependencies across fingers and hands. This work demonstrates and models significant inter- and intrahand relationships. Additionally, the developed Bayesian networks enable automated biometric identification system users to input their data to model finger variation for the computation of statistical conclusions. These relationships can be leveraged to predict pattern occurrences on other fingers which can be used to limit file penetration by filtering searches by finger position yielding increased search accuracy through a reduced search gallery.