APPLICATION OF DEEP LEARNING IN FINGER VEIN RECOGNITION

Abstract

Biometrics are body measurements and calculations related to human characteristics. Biometric identifiers are the distinctive and used to label and describe individuals. They are often related to the shape of the body such as face, fingerprint, palm veins, palm print, iris and so on. Biometric authentication is used in computer science to refer to security processes that verify a user's identity by biometrics. It is getting more popular in the past decade because of its reliability and convenience. Fingerprint and face recognition are the most used biometric techniques. However, there are a number of disadvantages of those technologies that people should take into consideration. For example, the age and occupation of a person or finger is sweaty or too dry may cause some difficulties in capturing a complete and accurate fingerprint image. Fingerprints are also relatively easy to be copied, so that usage may be restricted when used for security purposes. Therefore, finger vein recognition has attracted lots of attention as a new method for biometric authentication. Finger vein authentication is highly applicable, not only promising brings significant advantages but also solves the problems of other methods. Besides, Deep Learning has been strongly developed in recent years. It is considered as one of the most effective technologies in biomedical fields. Due to the potential benefits of Deep Learning and finger vein recognition, this study researches a method to recognize vein under the skin of finger using Deep Learning. Specifically, this research focuses on studying the principle of U-Net, which is a Convolutional Neural Network architecture, then applying U-net for detecting the vein pattern from the image. In other words, this research present Deep Learning approach that is to build a U-Net model that take a near-infrared image of the finger vein region as input, then the model performs semantic image segmentation on finger vein image, each pixel of the image will be labeled which class it belongs to, in this case the vein class and background class. By the end of this research, a U-net model should be built and trained, it should be able to detect vein patterns with 92% accuracy.