PERFORMANCE ANALYSIS OF FINGER VEIN RECOGNITION TECHNIQUE USING DEEP LEARNING

Abstract

Finger vein acknowledgment is a strategy for biometric confirmation that utilizations design acknowledgment procedures dependent on pictures of human finger vein designs underneath the skin's surface. Finger vein acknowledgment is utilized to recognize people and to confirm their character. Finger vein acknowledgment is a biometric validation framework that coordinates with the vascular example in a person's finger to recently got information. Hitachi created and protected a finger vein distinguishing proof framework in 2005. The innovation is essentially utilized for charge card validation, vehicle security, worker time and participation following, PC and organization confirmation, end point security and computerized teller machines. To acquire the example for the data set record, an individual embeds a finger into an attester terminal containing a close infrared light-emanating diode (LED) light and a monochrome charge-coupled gadget (CCD) camera. The haemoglobin in the blood assimilates close infrared LED light, which causes the vein framework to show up as a dim example of lines. The camera records the picture and the crude information is digitized and held in a data set of enrolled pictures. Vein designs are one of a kind to every person. Not at all like other biometric frameworks in any case, vein designs are practically difficult to fake since they are situated underneath the skin's surface and must be gotten from a living individual. Automated methods based on computer vision are being widely used for vein recognition. In this thesis, two novel methods for finger vein recognition are proposed. The first method is based on a hybrid filter. The second method is developed using deep learning techniques. As convolution neural networks have shown high efficiency in the field of computer vision. Thus, in the proposed method a Resnet18 model is used for the finger vein recognition. The proposed methods are applied on two publicly available databases. The results obtained are quite satisfactory and may be used for real life applications.