Publication: Recognition of person based on iris biometric using deep learning approach
Loading...
Date
2023-07
Authors
Looi, Wei Wen
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Iris biometric identification is a reliable and accurate method for verifying the identity of a person based on the unique patterns of their iris. However, different iris
databases may have different characteristics, such as image quality, resolution, and illumination, which can affect the performance of iris classifiers. In this paper, a deep learning approach is proposed, which focuses on the investigation on building a robust deep learning model for iris recognition classifier that can handle various databases with one model. A convolutional neural network (CNN) is developed to extract features from the iris images and classify them into different classes. The steps involved include pre-processing, localization, and segmentation of the iris images, which are essential for improving the accuracy and robustness of the system. In the proposed study of approach, the CNN architecture consists of 3 convolutional layers, 3 max-pooling layers, 2 fully connected layers, 1 flatten layer, 1 output layer, and is trained on a large dataset of iris images from 3 different sources, which are Code Ocean, IITD, and MMU, with a total of 4920 images. Each database is being trained for 30 runs based on number of epochs of 5, 10, 15, 20, and 25. Then the combinations of these databases are being trained. MMU database had an average of 70.909% accuracy. IITD database had an average of 92.560% accuracy. Code Ocean database had an average of 89.208% accuracy. The combinations of MMU and Code Ocean had an average of 85.473% accuracy. However, the combinations of all 3 databases only had an average of 0.203%. Limited data in a database will affect the model performance, as it limits CNN’s ability to capture all relevant features and leads to overfitting. Combinations of imbalance databases with one database dominates the overall dataset in terms of size or class distribution, the model may become biased towards the larger database and affects the performance of the model.