Fingerprint Pattern Classification Using Deep Learning, Singular Point Detection and Ridge Counting

Ho, Yik Wey (2026) Fingerprint Pattern Classification Using Deep Learning, Singular Point Detection and Ridge Counting. Final Year Project (Bachelor), Tunku Abdul Rahman University of Management and Technology.

Text HO YIK WEY_Full Text.pdf Restricted to Registered users only Download (9MB)

Abstract

This research explores the application of deep-learning techniques for fingerprint pattern classification and the hybrid of deep-learning method and the orientation-based method for the detection of core/delta singular point and the ridge count between the singular points. This study leverages Convolutional Neural Networks (CNN) integrated with advanced layers, including Residual Blocks and Kolmogorov–Arnold Network (KAN) layers, to enhance classification accuracy of fingerprint with various classes and the generalization capability of the deep-learning Model. Moreover, through the application of Optuna-based hyperparameter optimization and dataset balancing, the classification model is optimized for robustness, achieving reliable classification results on complex fingerprint dataset. The contribution of this research lies in its ability to improve fingerprint classification systems and the generalization capability of the deep-learning Model. The designed CNN Model in the project able to get test accuracy of 90.91% and the test loss of 0.23. The research introduces a hybrid methodology combining Faster R-CNN and orientation-based refinement algorithms that enable the accurate detection of core and delta singular points for the touchless fingerprint dataset. The performance of the Faster R-CNN alone gets the mAP of 0.7664 and the mean IoU of 0.7682 which is a reliable performance. Then, the hybrid method that consists of the Faster R-CNN and orientation-based method able to get the exact delta and core singular points with high reliable performance and accuracy. The detection method develops for the ridge count between the singular points also able to get the accurate ridge count of the touchless fingerprint image. The detection method for the singular points and ridge count between the singular points is robust to the noise and distortion in the touchless fingerprint and hence able to obtain the accurate result. This detection system is vital for forensic and biometric applications