Fuzzy extractors convert biometric data into reproducible uniform random strings, and make it possible to apply cryptographic techniques for biometric security. They are used to encrypt and authenticate user data with keys derived from biometric inputs. This research investigates how hardware security primitives can have provable cryptographic properties, a connection which is largely lacking in currently available hardware primitives. The development of such computational fuzzy extractors could result in substantially more efficient and reliable key extractors which may be better received by industry and other stakeholders, due to their improved efficiency and well-established security properties.
Computational fuzzy extractors derive keys from biometric sources including silicon biometric sources, and their security is based on the difficulty of problems such as Learning Parity With Noise (LPN). Existing computational fuzzy extractors require exponential time to extract keys when the bits generated by the biometric source contain a constant fraction of errors. The project explores the concept of a noise-avoiding trapdoor that results in a computational fuzzy extractor that can correct errors in polynomial time in a constant fraction of the bits generated by the biometric source. The security assumption is exactly the assumption of computational hardness of LPN. This approach remains secure under weaker assumptions about biometric data than previous schemes which assumed uniform distributions of biometric data. The project introduces high-school students to research in applied cryptography and security through the MIT PRIMES high-school outreach program.
