Biometric retina identification using artificial approach

Abstract

In this paper, we considered recognizing 2D retina pictures with a Convolutional Neural Network (CNN) for greater accuracy since retina-based identification is the most secure way of establishing identity and identifying people. An artificial neural network that is used to examine pixel input and recognize and process images is called a CNN. CNN algorithm has been selected to identify 2D retina images because through the CNN algorithm faster and better accuracy can be achieved. The retina identification process includes gray scaling of the RGB retina images, vessel extraction of the retina in the 2D images and then data augmentation is performed to increase datasets. Our method was evaluated on 3 databases- ARIA, DRIVE and STARE and we achieved test accuracy of 1 multiple times within 45 epochs. Test accuracy of 0.983 is received as the highest average accuracy among every 10 epochs. The implementation of the identification process was done using the PyTorch package.