A semi-supervised deep rule-based classifier for robust finger knuckle-print verification

Today, biometric recognition systems play an important role in various applications of different domains. Despite remarkable progress, their performance remains insufficient for security applications. Recently, semi-supervised deep rule classifier (SSDRB) is clearly explainable and universal classification tool used to solve different problems of classification or prediction. Thus, in this paper, we propose an effective scheme based SSDRB classifier for personal authentication systems, where, finger knuckle print (FKP) has been exploited. The proposed scheme is data driven and completely automatic. In this scheme, the pertinent and relevant features are extracted from the input finger knuckle image by binarized statistical image features descriptor (BSIF), which are then fed into fuzzy rules based multilayer semi-supervised learning approach based on a deep rule-based (DRB) classifier to decide whether the person is genuine or impostor. The experiments were conducted on the publicly available PolyU-FKP database provided by University of Hong Kong. The results are represented in form of rank-1, equal error rate (EER), cumulative match curve (CMC) and receiver operating characteristic (ROC) curves. The obtained results demonstrate that the proposed SSDRB classifier is a promising tool for the FKP biometric identification systems. Experimental results on the PolyU-FKP database show that the proposed SSDRB achieves lower error rates with an EER of 0.00% and a rank-1 of 99.90% on the FKP single modality outperforming several published methods.