Abstract
Personnel identification has become a mandatory requirement in a large number of applications extending from security to commercial nature in recent years. Identification mechanism using Biometric-based solutions has shown to overcome several drawbacks of traditional security measures. Among different biometric traits, fingerprint is one of the most universal, permanent and easy to acquire trait for personal identification. In this article, we investigate and evaluate the performance of the state-of-the-art machine learning algorithms employed in Minutiae based automatic fingerprint recognition. Fingerprint images from Public Domain Database (DB1) of FVC 2002 are used to carry out the experiments. Fingerprint image is initially preprocessed to enhance, binarize and skeletonize. Ridge ending and ridge bifurcation Minutiae features are then extracted and used for training and testing the Random Forest, Multilayer Perceptron, Radial Basis Functions and Naïve Bayesian machine learning Algorithms. A total of 80 instances and 150 attributes have been used in the experiments. The results show that Random Forest and Radial Basis Functions give better results for varying quality images compared to the other machine learning Algorithms and show the efficacy of these algorithms.