Ocular Biometrics: Human Recognition in Challenging Conditions

Abstract

The iris is the most reliable human biometric known to date. The iris texture has been shown to be effectively unique under ideal imaging conditions. However, as imaging constraints are relaxed, the iris becomes increasingly difficult to image, dra- matically decreasing its usability as an accurate biometric. Ocular recognition has been recently proposed as a way to complement iris recognition in less constrained imaging conditions. Ocular recognition refers to the use of additional information around the eye as part of the biometric information. This thesis proposes a definition for the term ocular region and shows how recognition performance using this region is more robust under challenging imaging conditions. It also proposes new approaches to ocular recognition that outperform iris recognition on challenging datasets, thus providing strong justification and motivation for further study of the ocular region as a biometric. These methods include an optimized scale invariant feature transform (SIFT) and a fusion method utilizing SIFT and Gabor filter encoding.