Title: Private Biometric Cryptosystems
Ph.D. Candidate: Chloe Cachet
Major Advisor: Dr. Benjamin Fuller
Committee Members: Dr. Ghada Almashaqbeh and Dr. Ariel Hamlin
Date/Time: Wednesday, June 21st, 2023, 12pm
Location: HBL Instruction 1102
Meeting link: https://uconn-cmr.webex.com/meet/csc18001
Biometrics are measurement of physical phenomena of the human body. They are used in a variety of applications for identification and authentication purposes.
By nature, biometrics cannot be renewed and compromises are definitive. This thesis aims to build systems that ensure privacy of the biometric templates.
For identification systems, where biometric records are usually collected and stored in large databases with few protections, this can be achieved with proximity searchable encryption. For authentication systems, this is typically achieved using fuzzy extractors that generate a stable cryptographic key from noisy values.
This talk will focus on the first part of the thesis which aims to build proximity searchable encryption for the iris biometric. Proximity searchable encryption handles proximity queries (finding all records within a bounded distance of the query point) over encrypted records. We propose an interactive proximity searchable encryption scheme that only leaks the size of the database, built from locality sensitive hashes (LSHs) and oblivious maps. To ensure high accuracy, the system searches for the disjunction of several LSHs of the biometric template. For the iris biometric this number of LSHs has to be high, approximately 1000. However for a specific query a lot of these LSHs will not match any stored value.
Our scheme is an adaptation of Boldyreva and Tang’s work on approximate k-nearest-neighbors search. We show that their system, designed for a small number of LSHs, cannot ensure both high accuracy and efficiency for a database of iris templates. We modify and optimize their design for a setting where most LSHs values do not match, as is the case for iris templates.