Secure Dynamic Password Generation from Dorsal Finger- Wrinkle Biometrics Using Deep Feature Embeddings and Fuzzy Extractors

Description

Passwords remain a dominant authentication factor but are repeatedly compromised through leaks,
phishing, and offline guessing. Building on a coding-based idea that converts dorsal (back-of-finger)
wrinkle images into machine-readable codes for automatic password creation, we propose a biometric–
cryptographic framework that (i) learns discriminative wrinkle embeddings using modern machine learning
(ML), (ii) derives a stable cryptographic key from noisy biometric measurements via a fuzzy extractor, and
(iii) produces rotating, short-lived passwords using standard one-time-password (OTP) constructions.
Unlike direct pixel-to-ASCII mappings and heuristic randomization, the proposed method separates
identity representation from password generation: ML provides repeatable feature vectors under realistic
acquisition variability, while cryptographic primitives provide revocability and stronger security properties
under database compromise and replay threats. We detail acquisition for four dorsal fingers, preprocessing
and ROI normalization, deep metric learning with Siamese/triplet objectives, multi-finger fusion, template
protection via cancelable transformations, fuzzy-extractor key derivation, and OTP synthesis with
HOTP/TOTP. Finally, we provide an IEEE-ready evaluation protocol and discuss limitations such as
liveness detection and presentation attacks.