This repository implements ClearEyes, the current state-of-the-art fuzzy key derivation system for iris biometrics that achieves 91% True Accept Rate while maintaining 105 bits of entropy. The implementation accompanies our research paper Fuzzy Extractors are Practical.
All results were evaluated on the IITD dataset.
The system combines three components:
- feap-fe: Feature extraction pipeline that transforms segmented iris images into binary feature vectors.
- lsh-lock: Analysis toolkit for zeta-sampling, quantifying entropy, and evaluating authentication performance
- sample-lock: Cryptographic implementation of the sample-then-lock fuzzy extractor written in C.
The params directory contains the essential configuration files needed to exactly reproduce our paper's results:
- Subsets for lsh-lock through zeta-sampling.
- Split for confidence and test subsets. Do not use confidence subsets for testing. Split was obtained with random selection.
- Input selection of templates used for enrollment and authentication.
In brief, to compute entropy and TAR:
cd lsh-lock
./target/release/lsh-lock analyze --templates /path/to/features/test/ --input /path/to/subsets/subset --dimensions 512 --count 250000
./target/release/lsh-lock tar-multi --templates /path/to/features/test --input /path/to/subsets/subset --count 250000 --dimensions 512 --base 3 --tries 5 --input-selection /path/to/base_3.jsonFor detailed build and usage instructions, see the README in each submodule.
If you use our work, please cite:
@misc{cryptoeprint:2024/100,
author = {Sohaib Ahmad and Sixia Chen and Luke Demarest and Benjamin Fuller and Caleb Manicke and Alexander Russell and Amey Shukla},
title = {Fuzzy Extractors are Practical: Cryptographic Strength Key Derivation from the Iris},
howpublished = {Cryptology {ePrint} Archive, Paper 2024/100},
year = {2024},
url = {https://eprint.iacr.org/2024/100}
}
This project is provided under GPL. The IITD dataset has its own licensing terms which must be respected.