An heuristic approach for the training dataset selection in fingerprint classification tasks

Abstract

Fingerprint classification is a key issue in automatic fingerprint identification systems. It aims to reduce the item search time within the fingerprint database without affecting the accuracy rate. In this paper an heuristic approach using only the directional image information for the training dataset selection in fingerprint classification tasks is described. The method combines a Fuzzy C-Means clustering method and a Naive Bayes Classifier and it is composed of three modules: the first module builds the working datasets, the second module extracts the training images dataset and, finally, the third module classifies fingerprint images in four classes. Unlike literature approaches using a lot of training examples, the proposed approach requires only 18 directional images per class. Experimental results, conducted on a consistent subset of the free downloadable PolyU database, show a classification rate of 87.59%.