Improved Finger Vein Recognition Using Generative Adversarial Network and Transfer Learning

The unique and secure nature of Finger Vein Recognition (FVR) has attracted considerable attention in the field of biometric authentication. However, challenges such as limited data availability and the complexity of vein patterns necessitate innovative approaches to improve recognition performance. A novel approach to FVR is proposed using Pix2Pix style Generative Adversarial Networks (GANs) for finger vein dataset augmentation providing synthetic images for training VGG16 model. The main objective of the paper is to enhance the quality and diversity of Finger Vein (FV) datasets while improving the robustness of the recognition system. Our approach involves training a GAN to generate realistic finger vein images that encapsulate the intricate vein patterns present in real-world scenarios. The study evaluates the proposed approach on the THU-FVFDT1 dataset, considering conventional augmentation and GAN-based augmentation. The results demonstrate that transfer learning from a conventionally augmented dataset to one augmented with GANs yields superior performance, as evidenced by the evaluation metrics. The experimental findings illustrate the effectiveness of dataset augmentation and transfer learning in the realm of finger vein recognition for biometric identification.