Personal identification using a cross-sectional hyperspectral image of a hand.

Significance: I explore hyperspectral imaging, a rapid and noninvasive technique with significant potential in biometrics and medical diagnosis. Personal identification was performed using cross-sectional hyperspectral images of palms, offering a simpler and more robust method than conventional vascular pattern identification methods.

Aim: I aim to demonstrate the potential of local cross-sectional hyperspectral palm images to identify individuals with high accuracy.

Approach: Hyperspectral imaging of palms, artificial intelligence (AI)-based region of interest (ROI) detection, feature vector extraction, and dimensionality reduction were utilized to validate personal identification accuracy using the area under the curve (AUC) and equal error rate (EER).

Results: The feature vectors extracted by the proposed method demonstrated higher intra-cluster similarity when the clustering data were reduced through uniform manifold approximation and projection compared with principal component analysis and -distributed stochastic neighbor embedding. A maximum AUC of 0.98 and an EER of 0.04% were observed.

Conclusions: I proposed a biometric method using cross-sectional hyperspectral imaging of human palms. The procedure includes AI-based ROI detection, feature extraction, dimension reduction, and intra- and inter-subject matching using Euclidean distances as a discriminant function. The proposed method has the potential to identify individuals with high accuracy.