Date of Graduation


Document Type


Degree Type



Statler College of Engineering and Mineral Resources


Lane Department of Computer Science and Electrical Engineering

Committee Chair

Thirimachos Bourlai

Committee Co-Chair

Bojan Cukic

Committee Member

Lawrence Hornak

Committee Member

Arun Ross

Committee Member

Kostas Sierros


Biometrics, or the science of identifying individuals based on their physiological or behavioral traits, has increasingly been used to replace typical identifying markers such as passwords, PIN numbers, passports, etc. Different modalities, such as face, fingerprint, iris, gait, etc. can be used for this purpose. One of the most studied forms of biometrics is face recognition (FR). Due to a number of advantages over typical visible to visible FR, recent trends have been pushing the FR community to perform cross-spectral matching of visible images to face images from higher spectra in the electromagnetic spectrum.;In this work, the SWIR band of the EM spectrum is the primary focus. Four main contributions relating to automatic eye detection and cross-spectral FR are discussed. First, a novel eye localization algorithm for the purpose of geometrically normalizing a face across multiple SWIR bands for FR algorithms is introduced. Using a template based scheme and a novel summation range filter, an extensive experimental analysis show that this algorithm is fast, robust, and highly accurate when compared to other available eye detection methods. Also, the eye locations produced by this algorithm provides higher FR results than all other tested approaches. This algorithm is then augmented and updated to quickly and accurately detect eyes in more challenging unconstrained datasets, spanning the EM spectrum. Additionally, a novel cross-spectral matching algorithm is introduced that attempts to bridge the gap between the visible and SWIR spectra. By fusing multiple photometric normalization combinations, the proposed algorithm is not only more efficient than other visible-SWIR matching algorithms, but more accurate in multiple challenging datasets. Finally, a novel pre-processing algorithm is discussed that bridges the gap between document (passport) and live face images. It is shown that the pre-processing scheme proposed, using inpainting and denoising techniques, significantly increases the cross-document face recognition performance.