Deep Learning-Enhanced Iris Biometrics: Integrating GAN-Generated Synthetic Data, Hybrid Feature Engineering, and Iterative Self-Learning for Robust and Scalable Recognition Systems
A.R. Kavitha, C. Sankari, T. Shalini · 2024
The iris, a distinctive anatomical characteristic of the human eye, functions as a biometric identifier even among identical twins with identical DNA. Conventional machine learning methods for iris detection frequently depended on manually crafted features, which were computationally demanding and produced inferior outcomes. Recent developments have transitioned to deep learning methodologies, utilizing robust structures for iris detection and segmentation to attain enhanced results. This study introduces a computational pipeline that emulates an iris recognition framework using synthetic data augmentation and feature extraction. The Proposed methodology employs Generative Adversarial Networks (GANs) to produce synthetic iris pictures, thereby enhancing the dataset to mitigate data constraints. The preprocessing procedures standardize the integrated actual and synthetic datasets, while feature extraction is executed by amalgamating mean pixel intensity computations with contextual information to produce hybrid feature vectors. The system emulates binary classification via a bespoke-trained model, attaining a classification accuracy of 98% in experimental conditions. Moreover, the pipeline incorporates a self-learning technique to progressively rectify misclassifications, guaranteeing adaptability and ongoing performance improvement. This comprehensive architecture illustrates the effectiveness of integrating synthetic data creation, deep feature extraction, and self-learning to attain dependable and scalable iris identification systems.