Fingerprint Recognition, Singularity Detection and Automated Ridge Density Analysis

Ong, Chun Chee (2026) Fingerprint Recognition, Singularity Detection and Automated Ridge Density Analysis. Final Year Project (Bachelor), Tunku Abdul Rahman University of Management and Technology.

Text Ong Chun Chee_Thesis.pdf Restricted to Registered users only Download (4MB)

Abstract

This research presents a dual-modality fingerprint recognition and analysis system designed to bridge the gap between traditional contact-based methods and modern touchless acquisition. By implementing a hybrid architecture by combining an EfficientNetV2-S backbone with a novel Kolmogorov-Arnold Network (KAN) head, the study achieved high classification accuracies of 89.93% for touchless and 87.84% for touch-based samples. Beyond classification, a Faster R-CNN framework with a ResNet-50 backbone was utilized for singularity detection, attaining a 91% testing accuracy in localizing core and delta landmarks. These structural landmarks facilitated an automated ridge-counting algorithm that demonstrated high forensic reliability, yielding a coefficient of determination (R2) of 0.9 and a Mean Absolute Error (MAE) of 1.02. The research journey revealed that data fidelity is paramount; transitioning from autolabelled to manually-verified datasets significantly resolved localization redundancy. While technical challenges such as overfitting and edge-case detection of delta points were encountered, they were mitigated through localized post-processing enhancement. The integration of KAN and targeted ridge-counting methodologies constitutes a significant contribution to biometric science, offering a hygienic, contactless framework that preserves the analytical depth of traditional forensics. Future work will focus on dataset diversification and model miniaturization for deployment in resource-constrained, real-time security environments