Robust and adaptive ROI extraction for hyperspectral dorsal hand vein images

Abstract

Hyperspectral dorsal hand vein image analysis for biometrics is a relatively new technology with great potential. Compared to traditional dorsal hand biometrics that use only one spectral band to capture and analyse the veins, hyperspectral imaging allows additional information to be included. Given the difficulties of processing hyperspectral dorsal hand images, including uneven illuminations, a noisy background, translation, and deformation, this study proposes a robust and adaptive region of interest (ROI) extraction algorithm. First, a novel knuckle pinky invariant point is located. Next, based on this invariant point a key line on the dorsal hand representing one side of the squared ROI is found. The remaining three sides of the ROI can then be formed. To evaluate the proposed method, both identification and verification experiments were conducted on a large hyperspectral dorsal hand vein database. The experimental results showed that the proposed method outperformed other dorsal hand ROI extraction algorithms for each hyperspectral band. Having performed well in both experiments, the groundwork has been laid to further analyse the extracted ROI in terms of feature extraction.