Band Selection for Improvement of Dorsal Hand Recognition

Abstract

In order to study the relationship between recognition performance of dorsal hand and spectra including visible and near-infrared light, a multispectral image capture system is established in this study, and it works well on the spectra with the wavelength from 520nm to 1040nm. The optimal band is considered to have the most effective features for recognition. Line feature extraction method is adopted regardless of the texture types. EER are calculated to reflect the differentiating ability across all bands after matching process. The spectrum of 880nm is testified to have the best performance of verification.