Threshold Analysis of Wavelet Based Fingerprint Feature Extraction Methods on Multiple Impression Dataset
P. Amoako-Yirenkyi *
Department of Mathematics, Kwame Nkrumah University of Science and Technology, Ghana and Scientific and Technical Computing, National Institute for Mathematical Sciences, Ghana.
N. K. Frempong
Department of Mathematics, Kwame Nkrumah University of Science and Technology, Ghana and Scientific and Technical Computing, National Institute for Mathematical Sciences, Ghana.
J. K. Appati
Department of Mathematics, Kwame Nkrumah University of Science and Technology, Ghana.
J. B. Hafron-Acquah
Department of Computer Science, Kwame Nkrumah University of Science and Technology, Ghana.
I. K. Dontwi
Department of Mathematics, Kwame Nkrumah University of Science and Technology, Ghana and Scientific and Technical Computing, National Institute for Mathematical Sciences, Ghana.
*Author to whom correspondence should be addressed.
Abstract
In recent years, fingerprint recognition has been moving through series of evolutions with the intent to decrease the False Acceptance Rate (FAR) and the False Rejection Rate (FRR) in order to achieve minimum Equal Error Rate (EER) while increasing recognition rate. In practical cases, fingerprint images stored in fingerprint databases may have come from scanners with different specifications under variant environmental conditions which may produce different or multiple impressions and backgrounds. The choice of what single and acceptable threshold value to use in order to characterize fingerprint features in images for recognition is therefore crucial in establishing a minimal EER. In this paper, we investigate and analyze the effect of several threshold values on EER when several families of wavelets based methods for feature extraction are applied on multiple impression datasets (Fingerprint Verification Competition-FVC2004). After conducting several threshold analysis on extracted features from multiple impression dataset, the results show that among the closely related wavelets families studied, the Reversed Bi-Orthogonal type 3:1 wavelet, analyzed with threshold value of 27 significantly topped with EER of 4:2% and a recognition rate of 95%. It however performed quite poorly outside of the threshold value indicating the importance of threshold analysis on datasets used for recognition.
Keywords: EER, FAR, FRR, Fingerprint recognition, Performance Rate, Recognition, Threshold Analysis, Wavelets.