Skip to main content
Thesis defences

PhD Oral Exam - Mohammed Ali, Electrical and Computer Engineering

A Novel Convolutional Neural Network Pore-Based Fingerprint Recognition System

Date & time
Friday, April 26, 2024
10 a.m. – 1 p.m.

This event is free


School of Graduate Studies


Nadeem Butt


Engineering, Computer Science and Visual Arts Integrated Complex
1515 St. Catherine W.
Room 005.251

Wheel chair accessible


When studying for a doctoral degree (PhD), candidates submit a thesis that provides a critical review of the current state of knowledge of the thesis subject as well as the student’s own contributions to the subject. The distinguishing criterion of doctoral graduate research is a significant and original contribution to knowledge.

Once accepted, the candidate presents the thesis orally. This oral exam is open to the public.


Biometrics play an important role in security measures, such as border control, forensics, business transactions and access to internet devices. For a biometric to be useful, it must have characteristics such as uniqueness, universality, permanence and collectability. These traits enable the system to leverage the distinct physical and behavioral attributes that are unique to each individual. There are a large number of biometrics, such as fingerprints, face, iris, gait and voice, that have been used in practice. Among these, the fingerprint is a notable biometric in view of its enduring nature, individual uniqueness and broad applicability. Fingerprint recognition systems traditionally rely on ridge patterns (Level 1) and minutiae (Level 2). However, these systems suffer from recognition accuracy with partial fingerprints. Level 3 features, such as pores, dots, and incipient ridges, offer the advantage of providing enhanced accuracy for fingerprint recognition with the availability of high-resolution fingerprint acquisition devices. Pores, with different sizes and shapes, situated alongside ridges, offer distinctive attributes crucial for individual identification, with studies indicating that a mere 20-40 pores are sufficient for establishing human identity, showing their superiority over minutiae even in a partial fingerprint segment. Moreover, in recent years, the use of convolutional neural networks (CNNs) for automatic feature extraction of biometrics has significantly improved the accuracy of biometric recognition systems.

A CNN-based pore fingerprint recognition system consists of two main modules, a pore detection module and a pore feature extraction and matching module. The pore detection module generates a pixel intensity map and uses it to determine the pore centroids. The pore feature extraction and matching module extracts relevant features of pores to generate pore representations from a query fingerprint and match them with those in a template fingerprint. In the existing CNN-based pore fingerprint recognition systems, CNN architectures are used for the extraction of pore features in both modules, in the first module for pore detection and in the second module for pore representations. However, these architectures lack in generating deep-level features that are sufficiently discriminative while maintaining computational efficiency. Moreover, the available knowledge on the fingerprint pores has not been taken into consideration optimally for the determination of pore centroids in the first module and metrics other than Euclidean distance have not been explored for matching of the pores in the second module.

The objective of this research is to develop a CNN-based pore fingerprint recognition scheme that is capable of providing a low-complexity and high-accuracy performance. For achieving this objective, the focus of this research is on the design of the two CNN architectures used in the two modules, the pore centroid determination scheme for the first module, and the pore matching scheme for the second module. The design of the CNN architecture of the two modules is aimed at generating features at different hierarchical levels in residual frameworks and fusing them to produce comprehensive sets of discriminative features for pore detection in the first CNN architecture and for pore representation in the second CNN architecture. Depthwise and depthwise separable convolution operations are judiciously used to keep the complexity of the networks low. In the proposed pore centroid determination scheme, the knowledge of the variation of the pore intensity from one region to another region of the fingerprint and the minimum distance between two neighboring pores are used, in addition to using the knowledge about the pore intensity, for detecting the pores. In the proposed pore matching scheme, a composite metric, encompassing the Euclidean distance, angle, and magnitudes difference between the vectors of pore representations, is proposed to measure the similarity between a pore in the query image and a pore in the template image.

Extensive experiments are performed on fingerprint images from the benchmark PolyU High-Resolution-Fingerprint dataset to demonstrate the effectiveness of the various strategies developed and used in the proposed scheme for fingerprint recognition.

Back to top

© Concordia University