Deep Learning for Iris Recognition: An Integration of Feature Extraction and Clustering

This paper introduces a comprehensive iris recognition system that integrates deep learning, clustering techniques, and a custom-designed dense layer architecture to address the challenges of iris detection and recognition. The CASIA-Iris-Thousand dataset is split into training and testing sets, enabling model evaluation. Preprocessing techniques, including image resizing and normalization, ensure standardized input. Feature extraction utilizes the VGG16 model to capture discriminative features from preprocessed iris images. Clustering with the K-means algorithm groups similar iris patterns, reducing variations and enhancing discriminative power. Additional layers, such as batch normalization, flatten, dense, dropout, and output layers, adapt the architecture for iris recognition and improve performance. Extensive experiments on the CASIA-Iris-Thousand dataset demonstrate the effectiveness of the approach, with the custom-designed dense layer architecture consistently surpassing conventional CNN architectures, leading to enhanced accuracy and robustness. The integration of clustering methods further improves identification of similarities and dissimilarities among iris patterns. This research contributes to a comprehensive iris recognition system, combining deep learning, clustering techniques, and a custom-designed dense layer architecture, with potential biometric authentication and surveillance applications. Privacy preservation is addressed, considering potential risks associated with iris-based identification. This study offers valuable insights for developing secure and privacy-aware iris recognition systems by integrating advanced techniques and emphasizing privacy.