Enhanced Fingerprint Alteration Detection Using Lightweight CNN Model Trained on SOCOFing Dataset

Fingerprint identification serves as a fundamental biometric authentication technique; yet, it is susceptible to potential vulnerabilities such as spoofing or manipulation. The identification of these modifications is crucial for ensuring the dependability of fingerprint-based verification. This research paper introduces a reliable methodology that utilises Convolutional Neural Networks (CNNs) that have been trained on the Sokoto Coventry Fingerprint Dataset (SOCOFing). SOCOFing is considered an optimal benchmark because to its inclusion of a wide range of fingerprint photos. The CNN model demonstrates a remarkable accuracy of 99.78% in accurately discerning between authentic and manipulated fingerprints. The use of hierarchical characteristics collected by convolutional layers is employed to identify small modifications. Data augmentation is used to improve the generalisation capability. The metrics of precision, recall, and F1-score provide an assessment of the model's efficacy in accurately identifying different types of alterations, such as partial modifications, picture splicing, and presentation assaults, with a high level of accuracy and resilience. This study places significant emphasis on the ongoing enhancement of fingerprint modification detection, which is of utmost importance given the constant evolution of spoofing tactics. The considerable precision shown by our Convolutional Neural Network (CNN) model indicates its viability for implementation in security-related domains, including but not limited to access control systems, border security measures, and criminal investigations.