Advanced Fingerprint Alteration Detection: A Comparative Analysis of Real and Synthetic Modifications Using InceptionV3 on the SOCOFing Dataset

This study investigates the efficacy of fingerprint alteration detection using advanced deep learning techniques, specifically focusing on both real and synthetically altered fingerprint images. Utilizing the Sokoto Coventry Fingerprint Dataset (SOCOFing), which comprises over 55,000 fingerprint images from 600 African subjects, we employed the Google InceptionV3 model to classify real and altered images under varying degrees of alteration complexity. Our experimental results demonstrate a robust performance of the model, achieving an accuracy of 91.04% for detecting alterations with easy alteration parameter settings, 98.07% for medium alteration parameter settings, and 96.47% for hard alteration parameter settings. The findings highlight the potential of deep learning frameworks like InceptionV3 in enhancing the reliability and security of biometric systems by effectively distinguishing between real and altered fingerprints, even under challenging conditions. This work contributes to the growing field of biometric spoof detection and emphasizes the need for resilient AI-based solutions in forensic and identity verification systems. Key Words: Fingerprint Alteration Detection, SOCOFing Dataset, InceptionV3, Synthetic Fingerprint Modifications, Deep Learning in Biometrics, Fingerprint Classification, Biometric Security, Altered Fingerprint Recognition, Transfer Learning, Convolutional Neural Networks (CNN)