Hand-dorsa Vein Recognition based on Deep Learning

Abstract

In last several years, deep learning methods have improved the performances of classification and recognition problems, especially for images. This paper investigates popular Convolutional Neural Networks (CNNs) on hand-dorsa vein recognition. To improve the performance of CNNs, a database enlargement method based on PCA reconstruction is proposed. To discuss the influence of dataset size, the enlarged dataset is sampled to form different datasets with the samples for each class are 50, 150 and 250 separately. Our method is run on the NCUT database and the enlarged database. Our method reaches the recognition rate of 99.61% when dataset size is 250 outperforming most other methods, meaning that the PCA reconstruction method is effective to improve the performance of CNNs.