Abstract
Fingerprint-based biometric systems are designed to authenticate and provide authorized access to the users in various security applications. However, such systems can be jeopardized by different kinds of presentation and spoof attacks by using different spoof materials. In this paper, an improved feature descriptor named as Quantized Fundamental Fingerprint Features (Q-FFF) is proposed for live fingerprint detection. The proposed feature descriptor integrates Level 1 and Level 3 features of fingerprints. Level 3 features of fingerprints such as ridge contours are extracted by the weighted linear combination (WLC) of magnitude of perceived stimuli and local contrast information in spatial domain. The novel magnitude of perceived stimuli with modified Weber law is proposed to compute the perceived stimuli of initial pixel intensities. The weights are explicitly determined by sigmoid function to interpolate the sparse information. Level 1 features such as ridge orientation is computed by retaining only the significant frequency components in frequency domain. Both ridge contours and ridge orientation are explicitly quantized into predefined intervals. The quantized feature sets are then integrated to populate a 2-D histogram. Performance evaluation of Q-FFF is determined on the three standard datasets of LivDet competition i.e. LivDet 2011, 2013 and 2015. Results comparison showed a reduction of Average Error Rates (AER) to 5.28, 2.0 and 4.62 on LivDet 2011, LivDet 2013 and LivDet 2015, respectively. (C) 2020 Elsevier Ltd. All rights reserved.