Fingerprint templates generated by the fractional Fourier, cosine and sine transforms and comparison of recognition accuracy and robustness

Abstract

Recently, I have proposed a new method for generating fingerprint templates using the discrete fractional Fourier transform (DFRT) in order to realize the optical fingerprint recognition system with the high recognition accuracy and the high robustness against attacks. In the previous study, I evaluated the recognition accuracy and robustness of the generated fingerprint templates by use of the fingerprint data used in the Fingerprint Verification Contest held in 2002 (FVC 2002). The fingerprint templates were generated not only by the DFRTs but the discrete fractional cosine transform (DFCT) and the discrete fractional sine transform (DFST). In the analyses, the fingerprint data to which Gaussian random noise was added were used to obtain the genuine distributions. In this study, first, the genuine fingerprint data in the FVC 2002 are used to obtain the genuine distributions correctly by performing the rotation and shift alignments. The recognition accuracy and robustness are also analyzed for the generated fingerprint templates by use of the equal error rate (EER) and the peak-signal to noise ratio (PSNR) between the original fingerprint image and the inverse-transformed image of the template, respectively. As a result, it is found that the most appropriate templates are the ADs of the DFCT and the DFST with a size of 96 by 96 pixels under the condition that the range of the transforms’ orders is between 0.1 and 1.0 because of the high recognition accuracy (EER is the order of 10 -6 %) and the high robustness (PSNR is order of several dB).