Neural Network Based Wrist Vein Identification Using Ordinary Camera

Abstract

Biometric access control systems find wide usage area nowadays and sensors cost pretty high. Wrist vein patterns are completely unique and visible through the skin therefore they are considerable as new biometric characteristics which can be used without specific sensors. In this study, wrist vein images were captured from 14 female and 20 male volunteers using an ordinary 5 MP mobile phone camera. Images captured approximately in distance 20 cm from top of volunteers' left wrists in daylight and 3 samples captured from each at different times. Clear wrist regions are windowed, FFT-based low-pass filters have been applied to image to suppress background noise. Sharpen filter and histogram equalization applied to images. Finally, to reduce processing time the images are resized 5 to 1 ratio. Three different databases obtained after conversion of images to arrays and feature reduction by PCA. RBF network, MLP and SVM classification structures used for wrist vein identification. Results proved to wrist vein identification without specific sensors are possible and proper to use in multimodal biometric systems.