Date of Award

2012

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Electrical Engineering

First Advisor

Kelly, John C.

Abstract

The Genetic & Evolutionary Computation (GEC) research community has seen the emergence of a new subarea, referred to as Genetic & Evolutionary Biometrics (GEB), as GECs have been applied to solve a variety of biometric problems. In this dissertation, we present three new GEB techniques for multibiometric recognition: Genetic & Evolutionary Feature Selection (GEFeS), Weighting (GEFeW), and Weighting/Selection (GEFeWS). Instead of selecting the most salient individual features, these techniques evolve subsets of the most salient combinations of features and/or weight features based on their discriminative ability in an effort to increase accuracy while decreasing the overall number of features needed for recognition. We also incorporate cross validation into our best performing technique in an attempt to evolve feature masks (FMs) that also generalize well to unseen subjects and we search the value preference space in an attempt to analyze its impact in respect to optimization and generalization. Our results show that by fusing the periocular biometric with the face, we can achieve higher recognition accuracies than using the two biometric modalities independently. Our results also show that our GEB techniques are able to achieve higher recognition rates than the baseline methods, while using significantly fewer features. In addition, by incorporating machine learning, we were able to create FMs that also generalize well to unseen subjects and use less than 50% of the extracted features. Finally, by searching the value preference space, we were able to determine which weights were most effective in terms of optimization and generalization.

Download

Share

COinS