Instant learning based on deep neural network with linear discriminant analysis features extraction for accurate iris recognition system

• Main Authors: Abdulhasan R.A., Abd Al-latief S.T., Kadhim S.M.
• Other Authors: 56905439800
• Format: Article
• Published: Springer 2025
• Subjects: Binary images; Biometrics; Database systems; Discriminant analysis; Economic and social effects; Extraction; Parameter estimation; Deep neural network deep neural network; Gaussian blur; Identity verification; Image contour; Iris recognition; Iris recognition systems; Linear discriminant analyze; Linear discriminate analyse linear discriminant analyse; Linear discriminate analysis; Shortest Processing Time; Deep neural networks

Biometric-based identity verification systems have gained substantial attention due to their ability to provide high-level security. Among these systems, iris recognition systems have emerged as one of the most accurate and complex verification approaches. However, an ideal recognition system with a short processing time has not yet been reported in the literature because of the trade-offs involved. In this article, a novel framework for an iris recognition system is proposed based on hybrid deep neural network (DNN) classification-based Linear Discriminant Analysis (LDA) for feature extraction. The developed system includes unique pre-processing steps for both training and testing datasets, which are modulated by greyscale conversation, Gaussian blurring, binary imaging, contour segmentation and resizing. The proposed LDA-DNN provides high accuracy and stability for human identity verification with a short processing time. The proposed model accomplishes this task perfectly without any loss, which is unique among this type of approach. The results are validated via five computed measurement parameters. Experimental results are obtained by applying the model to four typical existing databases for powerful validation. Moreover, the proposed LDA-DNN framework results are compared with outcome measures obtained for state-of-the-art iris recognition approaches. The experimental results illustrate the success and power of the proposed LDA-DNN model, which attains an accuracy of 100% within a time of 70 ms, corresponding to an ideal recognition result that validated using several databases. Furthermore, this work provides a model within a unique property, in which it does not require a specific database or measurement parameters for evaluation. ? The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature 2023.