A Deep Learning-Based Biometric System for Hand Dorsal Identification: Improving Accuracy and Robustness Through Tied-Rank Normalization

Biometric systems play a critical role in secure identity verification across various applications, yet existing modalities often face challenges related to hygiene, accuracy, and usability. This paper presents a novel hand dorsal biometric recognition system that leverages deep learning techniques and tied-rank normalization to enhance identification performance. Utilizing a dataset of 2,505 dorsal hand images from 500 individuals, our approach achieves an Equal Error Rate (EER) of 1.40% and a Recognition Rate (ROR) exceeding 94.8%. Performance evaluations through ROC and CMC analyses demonstrate the system's robustness and efficacy in distinguishing genuine users from imposters. Additionally, we explore the potential of multimodal fusion to further improve accuracy. Our findings indicate that hand dorsal biometrics is a promising solution for high-security environments, offering a contactless and hygienic alternative to traditional biometric systems.