Pose-Invariant Hand Geometry for Human Identification Using Feature Weighted k-NN Classifier

Abstract

Hand biometrics is globally deployed for automated human identification based on the discriminative geometric characteristics of hand. Advancements in hand biometric technologies are accomplished over several decades. The key objectives of this paper are two-fold. Firstly, it presents a comprehensive study on the state-of-the-art methods based on the hand images collected in an unconstraint environment. Secondly, a pose-invariant hand geometry system is excogitated. The experiments are conducted with the weighted geometric features computed from the fingers. The feature weighted k-nearest neighbor (fwk-NN) classifier is applied on the right- and left-hand images of the 500 subjects of the Bosphorus database for performance evaluation. The classification accuracy of 97% has been achieved for both of the hands using the fwk-NN classifier. Equal error rates (EER) of 5.94% and 6.08% are achieved for the right- and left-hand 500 subjects, respectively.