Optimized Score Level Fusion for Multi-Instance Finger Vein Recognition

The finger vein recognition system uses blood vessels inside the of an individual for identity verification. public is in favor a over conventional passwords or ID cards as biometric technology harder to forge, misplace, and share. In this study, histogram oriented gradients (HOG) features, which are robust against changes illumination position, extracted from personal recognition. To further increase amount information that can be used recognition, different instances vein, ranging index, middle, ring combined form multi-instance representation. This fusion approach preferred since it performed without requiring additional sensors feature extractors. combine effectively, score level adopted allow greater compatibility among wide range matches. Towards end, two methods proposed: Bayesian optimized support vector machine (SVM) (BSSF) SVM based (BSBF). results incrementally improved by optimizing hyperparameters HOG feature, matcher, weighted sum using optimization approach. considered kind knowledge-based takes into account previous attempts trials determine next trial, making efficient optimizer. By stratified cross-validation training process, proposed method able achieve lowest EER 0.48% 0.22% SDUMLA-HMT dataset UTFVP dataset, respectively.