Doctoral Dissertation Oral Defense
Title: Domain specific considerations and constructions for security applications of the iris biometric
Ph.D. Candidate: Luke Johnson
Major Advisors: Dr. Benjamin Fuller, Dr. Alexander Russell
Associate Advisors: Dr. Walter Krawec, Dr. Amir Herzberg
Date/Time: Monday, July 24th, 2023, 10:30 am
Location: HBL1102 and WebEx
Meeting link: https://uconn-cmr.webex.com/meet/lhj16101
Meeting number: 649 201 170
Despite decades of theoretical work, the problem of biometric authentication is yet to be solved because of its high error rate and negative results for general information theoretical solutions. Thus, domain specific considerations and computational assumptions are necessary. This dissertation presents both theoretical proofs of security and implementations.
A fuzzy extractor is a cryptographic primitive that allows for the production of stable randomness from noisy sources. The analytical metrics of a fuzzy extractor are efficiency, security, and correctness. Efficiency is the run time of the underlying algorithms. Security is measured in bits of entropy of the resulting randomness, and correctness is measured by the true accept rate of the authentication time algorithm. We present a number of results that will form the core of the thesis. These begin with a new fuzzy extractor that allows for the use of auxiliary confidence information at authentication time. We prove security for this new construction in the generic group model. Our next contribution shows how the use of global confidence information results in a new state-of-the-art fuzzy extractor. This scheme nearly doubles the amount of entropy estimated in the biometric feature vectors while maintaining a high true accept rate. Additionally, we present a novel information theoretic negative result for fuzzy extractors. We use a polynomial length advice string as a proxy for efficiency.