Fast identification of fingerprint

Abstract

The paper discusses a method of fast identification of fingerprints based on templates. The method is based on the properties of local neighborhoods of minutiae and consists of two stages. At the first stage, the neighborhoods of minutiae are compared independently from the request and reference templates. For each pair of minutiae, their similarity is estimated. To improve performance rating classes are introduced. They filter out unlikely pairs of minutiae. Some of the best ratings for each pair of minutiae of request and reference templates are accumulated in the histogram. As a result of the first stage, a histogram is constructed for the entire database. Based on the histogram, some of the most similar pairs of templates are selected. At the second stage, the already selected pairs of templates are compared additionally, taking into account the consolidation of minutiae. A significant increase in identification performance, in contrast to other publications, is achieved by eliminating false candidates in the first stage.