Abstract:
Latelyidentitymanagementhasbeenfullydigitizedduetowhichtheworkonbiometrics has gained significant popularity. An increase in the adoption of biometrics in identity management means that sensitive data of the individuals can be easily compromised if unauthorized access is granted to the systems. This can lead to legal issues as it leads to an infringement of privacy of the people. In this context, the use of tokens, personal identification numbers and passwords is deemed inadequate for protecting the personal data of the people. This has stimulated research in domain of template protection and use of various methods has been adopted for protecting the biometric templates of the individuals. In this thesis, we have adopted ‘Cancelable Biometrics’ for template protectionusingtrans form based methods. WehaveusedGaborfilteringpriortofeature extraction as it is a non-invertible transform and satisfies one of the key requirements (non-invertibility) of the templates. Gabor filtering is followed by feature extraction using the VGG19 network trained using ‘imagenet’ weights. The templates obtained still are not able to satisfy a key requirement (unlinkability) which is achieved by using random projection performed for every user using a ‘key’. If compromised, the existing template can be revoked and a new template can be obtained using another key, leading toauniquesetoffeaturesforthesameuser. Wehaveperformedourexperimentsonthe Multispectral Palm Image Dataset (CASIA). Our experiments show that the proposed method outperforms the other methods that have been considered in this thesis. The method has been analyzed with respect to privacy with theoretical evidence for nonorthogonalityofthetemplates,andempiricalevidenceforunlinkabilityusingtheT-test.
