Craig I. Watson

In this report, we propose a new definition of quality of fingerprint impressions and present detailed algorithms to measure image quality for fingerprints. We define fingerprint image quality as a predictor of matcher performance before a matcher algorithm is applied. This means presenting the matcher with good quality fingerprint images will result in high matcher performance, and vice versa, the matcher will perform poorly for poor quality fingerprints. We also have carried out an objective evaluation of the quality assessment of fingerprint images. Our quality measure is implemented in the C programming language and has been tested on 20 different live scan and paper fingerprints datasets collected in different operational settings. Our implementation is publicly, but export controlled, available as part of NIST's fingerprint software distribution.

This report describes a system we have developed that automatically classifies images of fingerprints into six pattern- level classes. Automatic classification is useful in an Automated Fingerprint Identification System (AFIS) because it can be used to partition the database of fingerprint cards and thereby reduce the amount of work that must be performed by the fingerprint matcher. Our program takes gray-level images of fingerprints as input, and for each fingerprint it produces a hypothesized classification as arch, left loop, right loop, scar, tented arch, or whorl, as well as a. number indicating how much confidence should be assigned to its classification decision. The system performs these processing steps: image segmentation; image enhancement; feature extraction; registration; application of a linear transform that both applies a pattern of regional weights and reduces dimensionality; running of a main classifier, which is a Probabilistic Neural Net, and of an auxiliary whorl-detecting classifier that traces and analyzes pseudoridges (approximate trajectories through the ridge flow); and finally, the combining of the outputs of the main and auxiliary classifiers so as to decide on a hypothesized class and a confidence level. The program's memory and disk space requirements can be met by a typical desktop workstation. The distribution consists of source code, data files, a demonstration set of 2700 fingerprint images, and documentation.

This document provides guidance on how the NIST Biometric Image Software (NBIS) nonexport controlled packages are installed and executed.Its content and format is one of user's guide and reference manual.Some algorithmic overview is provided, but more details can be found in the cited references.The Table of Contents provides the reader a map into the document, and the hyperlinks in the electronic version enable the reader to effectively navigate the document and locate desired information.These hyperlinks are unavailable when using a paper copy of the document.