The privacy/security tradeoff across jointly designed linear authentication systems

In the area of secure biometrics, work has been done to build an information theoretic framework characterizing privacy and security of single biometric systems. People have worked extensively on designing such systems, some cryptographic in nature, and others tied to error correcting codes. However, there is still little known about security and privacy across multiple jointly designed systems. This work will focus on the privacy/security tradeoff across multiple “secure sketch” biometric systems. Secure sketch is a type of biometric system architecture related to error-correcting codes where a system is characterized by a parity-check matrix over a finite field, or equivalently by a subspace of a vector space over that same field. Given a set of systems (a design), we introduce worst-case measures of privacy leakage and security in the case that a subset of the systems becomes compromised. It turns out that more secure designs are necessarily less private and vice versa. We study the tradeoff between privacy and security by relaxing a restricted version of the problem, by studying the algebraic structure of the problem, and by formulating graph theoretic questions. These approaches generate bounds on achievable privacy/security pairs.