Iris Recognition System Based On Efficient Model For CNN Features Extraction And SVM Classifier

Abstract

Iris recognition relied on iris features to verify and identify a person's identification. The iris texture provides a number of advantages, including long-term stability, ease of use, and excellent recognition accuracy. The most critical stage in the iris recognition system is extracting effective features. The majority of earlier iris recognition feature extraction methods used hand-crafted features. The Convolutional Neural Network (CNN) has recently been shown to be effective in most recognition problems. As a result, using CNN-extracted features in an iris identification system attracts our attention. In this paper, we will present the details of building an efficient system for iris recognition based on a number of design stages. Firstly, the pre-processing techniques, segmentation by using Circular Hough Transform (CHT), and Normalization by using Duagman Rubber-Sheet Model are applied to the image of the human eye to determine the region of the iris. Secondly, the new structure of CNN architecture is proposed to extract features from the normalized iris region. Finally, the SVM classifier is applied to classify the images in the CASIA- Iris Dataset V1. The hyper-parameters chosen and the deep networks and optimizers tuned to affect the performance of our proposed system. For the dataset utilized in the experiments, the suggested CNN architecture achieves 99.07% accuracy, outperforming current approaches.