Towards Privacy-Preserving Comparison of Finger Vein Patterns
Beneficiary: Marta Gomez-Barrero, PhD Student at ATVS, Universidad Autonoma de Madrid, Spain
Host: Prof. Dr. Christoph Busch, NISlab, Gjovik Universitiy College, Norway
Period: 01/01/2015 – 31/03/2015
Purpose of the STSM
Even though biometric recognition systems are nowadays widely deployed, many concerns arise from the privacy issues posed by biometric data storage. Several approaches, based on irreversible transformations or combining cryptographic techniques with biometric recognition have been proposed. Among them, Bloom filter based template protection has been successfully applied to different biometric traits (iris, face, fingerprint) granting irreversibility while preserving verification accuracy.
In the present this Short Time Scientific Mission (STSM), the Bloom filter based approach was applied to finger vein data and its fusion with fingerprint information. The performance of both the unimodal and the multimodal systems has been assessed, as well as the irreversibility provided.
Work carried out during the STSM
- Development of a biometric template protection system based on finger vein patterns and Bloom filters. A similar approach to those of face and iris is followed: the initial binary template, based on Maximum Curvature Points, is divided into nBlocks rectangular blocks, and one Bloom filter is extracted from each of them.
- Integration with the previous work on fingerprints and the multimodal finger sensor.
- Development of multimodal biometric template protection system based on a feature level fusion of finger vein patterns and fingerprints, protected with Bloom filters. This is a user-friendly combination, since both traits can be acquired using the same sensor. Unlike in previous works, the protected templates now have different sizes.
Results of the STSM
In order to validate the developed systems, the accuracy and irreversibility of the final protected systems will be carefully analysed: is the recognition performance maintained after applying the privacy layer, i.e., Bloom filter transformation? Is it practically infeasible for an eventual attacker to infer any biometric data from the secured templates? Should the previous two objectives be reached, it could be argued that Bloom filters provide an efficient and general method to protect fixed length biometric templates as it has been shown for iris, face, fingerprints and vein patterns.
Experiments were carried out on the multimodal SDUMLA-HMT DB, comprising face, iris, finger vein, fingerprint and gait data from 106 subjects. The Detection Error Trade-Off (DET) curves show that performance is even improved with the addition of the Bloom filter based template protection scheme to the fingervein data.
Regarding the multibiometric scenario, performance is improved with respect to the unimodal protection systems. On the other hand, even if it decreases with respect to the fused unprotected scenario, it should be taken into account that the privacy and security of the subject is not only increased by the Bloom filter scheme: while in the unprotected scenario, fusion is carried out in the score level, a feature level fusion is implemented in the protected system.
In both cases, we can conclude that templates are also irreversible due to average number of bits activated in each Bloom filter.
As shared future research lines, we could highlight:
- The combination of the Bloom filter based template protection scheme with another protection approach based on honey templates, developed at the NISlab, which further protects the privacy of the subject.
- The development of a general framework for the application of Bloom filters to any given biometric characteristic.