Abstract

Fingerprint recognition is the most employed biometric method for identification and verification purposes. Fingerprint images are classified into five categories according to the morphology of their ridges, which decreases the database penetration rate on an identification scheme. The classification procedure mainly starts with the feature extraction from the fingerprint sample, based on minutiae obtained from terminations and bifurcations of ridges. Afterward, the classification process is usually carried out by some artificial neural networks under supervised learning. Recently, convolutional neural networks are utilized as a potential alternative, by showing accuracies close to 100% with a high cost of learning times even using high-performance computing. On the other hand, the extreme learning machine (ELM) has emerged as a novel algorithm for the single-hidden layer feed-forward neural network, because of its good generalization performance at extremely fast learning speed. In this work, we introduce the ELMs for the fingerprint classification problem. The superior ELMs are given by the mapping activation function and the number of hidden nodes that maximize the accuracy of the classification; a heuristic approach is carried out to find these parameters. The studied databases are the NIST-4 and SFINGE, which are composed by different quantity and quality of fingerprint samples. Results show that ELM classification by using the feature descriptor of Hong08 achieves very high accuracy and low penetration rate, reducing severally the training time in comparison with deep learning approaches.