A Robust Wavelet-based Approach to Fingerprint Indentification

Abstract

A robust fingerprint recognition system based on marginal statistics of 2D wavelet transform is introduced which significantly improves the accuracy of previous wavelet based approaches due to 1) a better selection of features extracted from the wavelet transform, and 2) a more accurate distance measure to find the similarity between fingerprints. A combination of Jain and Poincare algorithms is employed to locate the fingerprint reference point. The main part of the fingerprint image is chosen as a rectangle with the reference point at its center. The image is then divided into nonoverlapping sub-images, the wavelet transform is applied to the bi-level sub-images, and Generalized Gaussian Density (GGD) features are extracted from each wavelet sub band. Finally, the fingerprint recognition is done through the k-Nearest Neighbor (k-NN) classification employing Kullback-Leibler Distance (KLD) measure. Our test results confirm the superiority of the proposed method over the current fingerprint recognition methods.