Abstract: Fingerprint-based biometric systems are vulnerable to attacks involving altered or forged fingerprints. This paper introduces a robust machine learning model for detecting altered fingerprints, utilizing the SOCOfing dataset containing 6,000 real and 49,270 altered fingerprint images. The model employs a convolutional neural network (CNN) to extract critical features such as ridge patterns, minutiae points, and texture details, achieving high accuracy and reliability. Our results demonstrate significant improvements in biometric security, paving the way for advanced applications in forensics and authentication systems. The findings also highlight the importance of AI-based security solutions and propose methods to scale the model for real-world applications. Future studies can focus on optimizing CNN architectures and integrating hybrid models for increased robustness.
Keywords: Altered Fingerprints, Machine Learning, SOCOfing Dataset, Convolutional Neural Networks, Biometric Security
|
DOI:
10.17148/IJARCCE.2024.131243
