An Improved Residual Pix2Pix Generative Adversarial Network for Finger Vein Segmentation

Abstract

Today, biometric technology has become a reliable solution for secure authentication. The uniqueness of finger vein patterns, contactless data collection, and low implementation costs make finger vein (FV) recognition stand out among other biometric methods. Researchers have developed many techniques to accurately recognize and classify vein patterns in fingers. In this work, Generative Adversarial Networks (GANs) are used to create a precise finger vein prediction model. The results highlight the effectiveness of the ResNet50 PIX2PIX Generative Adversarial Network and its potential applications in biometric recognition systems. Images are sourced from the THU-FVFDT1 dataset and the Kaggle dataset. The process begins with standardizing the data and removing noise by using the Noise2Void model and pre-processing the images, followed by GAN-based segmentation to generate Segmented masks from the pre-processed images. This approach produced good results for finger vein recognition. An Akaze Detectoris used for feature matching. The performance of the ResNet50 Pix2Pix GAN achieves an accuracy of 98.73%, which is greater than kaggle dataset. These findings demonstrate the effectiveness of ResNet50 Pix2Pix GAN in improving finger vein segmentation and its potential to enhance biometric authentication and identification industries.