Bayes Classification for the Fingerprint Retrieval

Abstract

The Fingerprint is the most commonly used biometric property in security, commerce, industrial, civilian and forensic applications. The goal is to raise the recognition rate in the fingerprint retrieval system. In this work, the Bayes classifier is adopted assuming Gaussian statistics. The set of training samples are expanded by spatial modeling technique and implement a variant of the Fisher’s Linear Discriminant Analysis (FLDA) for dimension reduction and Quadratic Discriminant Analysis (QDA) for lowering estimation errors. Finally calculating the probabilistic features for Gabor and Minutiae which helps to reduce the error rate about 75% which outperforms the K-NN classifier where the error rate was about 30-60%. The accuracy and Speed are evaluated using FVC2004 database and satisfactory retrieval performance is achieved. Thus the objective of the Fingerprint Retrieval system that is efficient and accurate is build.

Keywords: Fingerprint retrieval, Bayes classifier, Gaussian statistics, training samples, Fishers Linear Discriminant Analysis (FLDA), Quadratic Discriminant Analysis (QDA), minutiae, FVC database.