Author ORCID Identifier

https://orcid.org/0000-0002-2794-2351

Semester

Fall

Date of Graduation

2025

Document Type

Dissertation

Degree Type

PhD

College

Statler College of Engineering and Mineral Resources

Department

Lane Department of Computer Science and Electrical Engineering

Committee Chair

Jeremy Dawson

Committee Member

Brian Powell

Committee Member

Gianfranco Doretto

Committee Member

Jacqueline Speir

Committee Member

Yenumula Reddy

Abstract

Fingerprint-based biometric recognition remains one of the most dependable and widely adopted approaches for identity verification due to its permanence and distinctiveness. Recent advancements in mobile and contactless imaging have extended fingerprint acquisition beyond controlled environments into unconstrained, real-world conditions through fingerphotos, contactless fingerprint images captured by digital or smartphone cameras. While this paradigm shift enhances accessibility and user convenience, it introduces significant technical challenges. Variations in illumination, focus, and motion blur often degrade ridge patterns, making accurate feature extraction and matching more difficult. Similarly, in forensic applications, latent fingerprints, incomplete or smudged prints lifted from surfaces, pose unique challenges due to their poor ridge quality and scarcity of labeled data. At the core of this research is the design and analysis of deep generative and self-supervised architectures tailored for fingerprint data. Conditional and multi-stage Generative Adversarial Networks (GANs) are investigated for their ability to restore blurred or degraded fingerphotos, incorporating ridge extraction and identity-verification subnetworks to maintain fine-grained fingerprint information. Attention guided mechanisms are examined to prioritize severely distorted regions during reconstruction, enabling more accurate ridge recovery under complex degradations. To mitigate the limited availability of latent fingerprints, a style transfer-based synthesis framework is proposed to generate realistic latent samples by blending clean ridge structures with latent-style textures. In parallel, a self-supervised dual-encoder framework is developed for fingerphoto quality assessment, learning to predict quality measures aligned with recognition utility without manual annotation. These studies collectively analyze how generative and representation learning models can be adapted to the structural and statistical characteristics of fingerprint data. Together, these deep learning-based methods directly address the principal obstacles in fingerphoto and latent fingerprint recognition: image degradation, data scarcity, and image quality. The restoration models improve the fidelity of ridge patterns in degraded fingerphotos, facilitating reliable minutiae extraction and matching. The synthesis framework expands the diversity of latent fingerprint datasets, supporting robust training and performance evaluation of modern recognition algorithms. The self-supervised quality assessment model provides an automatic mechanism to filter or guide capture processes based on predicted matching utility. By integrating these complementary approaches, we establish a unified deep learning framework for optimizing the reliability, efficiency, and overall performance of contactless and latent fingerprint biometric systems.

Download

Share

COinS