Massively Parallel Neural Network Fingerprint Classification System

摘要

The report describes a massively parallel fingerprint classification systemdeveloped at the National Institute of Standards and Technology (NIST) for the FBI that uses image-based ridge-valley features, K-L transforms, and neural networks to perform pattern level classification. The speed of classification is 0.54 seconds per fingerprint on a massively parallel computer. The system is capable of 88% classification accuracy with 10% rejects. As part of the development activity a sample of 4000 fingerprints, 2000 matched pairs, was collected and publicly released as NIST Special Database 4.