Idiap researchers develop vein biometric template protection approach with biohashing
A pair of methods for protecting vein biometrics templates from vulnerability to theft or leakage without degrading recognition performance have been developed by a pair of researchers from the Idiap Research Institute.
Sebastien Marcel and Hatesh Otroshi researched the application of deep neural networks for biometric template protection with funding from the TReSPAsS-ETN (for training in secure and privacy-preserving biometrics, early training networks) project of the EU Marie Sklodowska-Curie ITN-European Industrial Doctorate (EID) program.
‘Deep auto-encoding and biohashing for secure finger vein recognition’ was published by the Proceedings of the 2021 IEEE International Conference on Acoustics, Speech, and Signal Processing, while the more extensive ‘Towards protecting and enhancing vascular biometric recognition methods via biohashing and deep neural networks’was published in the journal IEEE Transactions on Biometrics, Behavior, and Identity Science.
The approach involves protecting vein templates by biohashing deep biometric features. The researchers captured the deep features through “a convolutional auto-encoder neural network with a multi-term loss function.” The dimension of features was reduced with a deep neural network, and then extracted the features with the auto-encoder at the embedding layer.
Otroshi and Marcel believe their proposal to reduce the dimensionality of the vein images with a deep auto-encoder before biohashing is novel in the field.
The former paper proposes an auto-encoder (AE) for feature extraction, and compares the matching performance with the resulting templates in normal scenarios and situations in which the user’s private key (however unlikely in practice) has been leaked against Wide Line Detector (WLD), Repeated Line Tracking (RLT) and Maximum Curvature (MC) feature extraction algorithms with biohashed templates. The second paper tests each of the above extraction algorithms alone, in combination with biohashing, those two methods combined separately with Principal Component Analysis (PCA) and with the AE extraction algorithm.
The combined AE-plus-biohash approach yields superior results in the researcher’s simulation to those without the AE in both the normal and stolen scenarios.
Although the research primarily focussed on finger vein biometrics, testing for palm and wrist vein biometrics showed similarly promising results.
A market report last year forecast revenues from vein biometrics to reach $1 billion a year by 2029.