Robust Fingerprint Identification System Using Backpropagation and ART Neural Networks
A.A. Ilumoka, M. Talluri, M. Eng
University of Hartford, US
biometric systems, fingerprint recognition, neural networks, artificial intelligence
This paper describes a robust minutiae-based fingerprint identification method suitable for use in small populations. System employs two serially connected neural networks in which fingerprint feature extraction is carried out by the first network – a backpropagation neural network and matching by the second - an adaptive resonance theory network which performs the decision making task of matching acquired fingerprint to templates in a database. The approach has been applied to a real database of noisy fingerprints derived from the 2002 Fingerprint Verification Competition (FVC2002) and has achieved error rates as low as 4% at penetration rates of 100%.
Nanotech 2008 Conference Program Abstract