Dorsal hand Vein Authentication System Using Convolution Neural Network

Abstract

This study proposes using a dorsal hand vein authentication system using transfer learning from convolutional neural network models of VGG16 and VGG19. The required images were obtained from Bosphorus Hand Vein Database. Among the 100 users, the first 80 users were treated as registered users, while the remaining users as unregistered users. 960 left-hand images of the registered users were trained during the training phase. Meanwhile, 100 images, consisting of 80 registered and 20 unregistered users, were randomly selected for the authentication application testing. Our results showed that VGG19 produced a superior validation accuracy as compared to that of VGG16 given by 96.9% and 94.3%, respectively. The testing accuracy of VGG16 and VGG19 is 99% and 100%, respectively. Since VGG19 is shown to outperform its shallower counterpart, we implemented a User Interface (UI) based on VGG19 model for dorsal hand vein identification. These findings indicate that our system may be deployed for biometric authentication in the future for a more efficient and secure implementation of person identification and imposter detection