Finger Vein Recognition Using Deep Learning

Abstract

Recently, finger vein recognition technology has gained wide acceptance in both research as well as in commercial uses like access control and authentication. Finger vein recognition is a novel biometric technology which is challenging to spoof and has a wide array of potential applications. Many deep learning based finger vein recognition system has been proposed so far. The vein images are always prone to quality degradations due to the noise, blurring and illumination variations introduced while capturing the images using near infrared technique. However, most of the existing deep learning methods for finger vein recognition are based on ideal vein images (images with minimum image quality). Since the network is trained based on the vein images which are having a minimum image quality, the performance of recognition may get affected when the vein image quality is poor. We propose a transfer learning based model which is trained using vein images with varying image quality. From a unique vein image, four different quality images (original image quality, blurred image, noisy image1, and noisy image2) will be generated and used for training the model. To the best of our knowledge, this is the first work based on transfer learning model that relies upon varying qualities of vein images in order to improve the overall recognition performance. We have utilized SDUMLA vein image dataset for experiments. The experimental results shows that the proposed approach can perform better than the existing deep learning based methods.