Finger knuckle pattern person identification system based on LDP-NPE and machine learning methods

Biometric-based individual distinguishing proof is a successful strategy for consequently perceiving, with high certainty, an individual's character. The utilization of finger knuckle pictures for individual ID has shown promising outcomes and produced a ton of interest in biometrics. By seeing that the surface example delivered by twisting the finger knuckle is profoundly particular, in this paper we present a new biometric validation framework utilizing finger-knuckle-print (FKP) imaging. In this paper, another methodology in view of neighborhood surface examples is proposed. Local derivative pattern (LDP) histogram is investigated for FKP description. Then based on neighborhood preserving embedding (NPE) is used for dimension reduction to the feature vector. The feature extraction method is computed and evaluated in the identification framework task. The machine learning methods (multiclass support vector machine (MSVM), random forest (RF), k-nearest neighbor (KNN)) are proposed for classification. The system is tested on the PolyU finger knuckle database. The empirical results proved that the proposed model has the best results with RF. Moreover, our proposed LDP-NPE model has been evaluated and the results show remarkable efficiency in comparison with previous work. Experimentally, the proposed model has better accuracy as reflected by 99.65%.