Modular Decomposition of Fingerprint Time Series Captures for the Liveness Check
Aditya Abhyankar, Stephanie Schuckers · International Journal of Computer and Electrical Engineering · 2010
This work demonstrates a faster approach for liveness detection in fingerprint devices. The physiological phenomenon of perspiration, observed in time-series fingerprint images of live people, is used as a measure to classify `live' fingers fromnot live' fingers. Pre-processing involves finding the singularity points using wavelets in the fingerprint images and transforming the information back in the spatial domain to form a spatial domain signal. Wavelet packet sieving is used to tune the modes so as to gain physical significance with reference to the evolving perspiration pattern inlive' fingers. The percentage of energy contribution in the difference modes is used as a measure to differentiate live fingers from others. The proposed algorithm was applied to a data set of approximately 58 live, 50 spoof and 28 cadaver fingerprint images captured at 0 and after 2 sec, from three different types of scanners. An overall classification rate of $93.7 \%$ was achieved across all the three scanners. sufficient number of singularity points is obtained, only then, the obtained signal is subjected to Hilbert Huang Technique. Wavelet packet sieving is used to tune the modes so as to gain physical significance with reference to the evolving perspiration pattern inlive' fingers. The percentage of the energy contribution in the difference modes is used as a quantified measure to differentiate live fingers from others. It was observed that the singularity points act asquality check' points and hence bad quality cadaver images and partial spoof images were rejected due to number of singularity points below the selected threshold, before further processing. In this paper, section 2 presents data management. Section 3 describes the singularity point detection part of the algorithm. Section 4 gives the background for empirical mode decomposition (EMD), Hilbert-Huang transform (HHT) and wavelet packet transform (WPT). Section 5 presents the overall algorithm and sections 6 and 7 present results and conclusion.