Robust and Scalable Fingerprint Recognition Leveraging Data Augmentation and Capsule Network
Dr.AB.Hajira Be1, N.Gayathri2
1 Associate Professor
Department of Computer Applications
Karpaga Vinayaga College of Engineering and Technology
Maduranthagam TK
2PG Student
Department of Computer Applications
Karpaga Vinayaga College of Engineering and Technology
*Corresponding Author: Gayathri D Email: gayathri7555@gmail.com
Abstract - Fingerprint recognition is a pivotal technology in modern biometric systems, offering robust and secure methods for identity verification and authentication. This project presents a comprehensive approach to fingerprint recognition by leveraging advanced deep learning techniques, particularly capsule networks, alongside classical image augmentation methods to enhance the robustness and accuracy of the system. The system comprises three core modules: Data Augmentation, Model Training, and Testing and Prediction. The augmentation module applies a diverse range of transformations, including color filtering, grayscale conversion, and blur effects, to generate a wide variety of training samples, thereby improving the model's ability to generalize across different fingerprint patterns. These augmented datasets are encapsulated into compact zip files for efficient storage and retrieval. The training module employs a user-selected deep learning architecture, such as ResNet50, VGG16, or EfficientNetB0, as a base model, with a capsule network layer added for capturing spatial hierarchies within the fingerprint features. This network is designed to classify fingerprints into distinct categories with high accuracy, utilizing a categorical cross-entropy loss function optimized by Adam. The training pipeline includes data generators for loading and preprocessing fingerprint images, ensuring consistency and scalability. In the testing module, a trained model is deployed to predict fingerprint classes from unseen images. A graphical interface guides users through selecting pre-trained models and testing images, while prediction results, including confidence scores, are visually presented alongside the input images. This seamless integration enhances user interaction and interpretability of the results. The project emphasizes modularity and user customization, allowing the selection of models, adjustment of parameters, and intuitive augmentation controls. By combining traditional image processing techniques with state-of-the-art deep learning architectures, this system demonstrates the potential for accurate, efficient, and user-friendly fingerprint recognition solutions, suitable for a wide range of applications in security and identity management.
Key Words: Data Augmentation, Genetic Operators, IoT, Fingerprinting, Sigfox, Localization, Positioning, Reproducibility, Machine Learning, knn.
