Robust cancellable biometrics scheme based on neural networks

Abstract

Several cancellable biometrics (CBs) techniques have been proposed to protect biometric data and maintain users' privacy. Although such techniques can withstand brute-force and/or pre-image attacks, they are vulnerable to correlation attacks. In this study, the authors propose a novel correlation attack-resistant CBs scheme that is based on a convolution operation and a bidirectional associative memory (BAM) neural network. The proposed scheme utilises BAM to bind biometric templates to random bit-strings in a secure and efficient manner. These random bit-strings are then employed to derive cancellable templates from the true templates linked to them via BAM weights, which are safely stored with the generated cancellable template in the system database. In this study, linear convolution is adopted as the cancellable transformation process. The result of convolving the original biometric template with the transformation key is binarised according to a predefined threshold to thwart blind de-convolution. The security of the proposed scheme against different attacks is analysed and experiments on the CASIA-IrisV3-Interval dataset illustrate the efficacy of the proposed scheme.

Secure Triplet Loss for End-to-End Deep Biometrics

“…Besides accounting for trait variability, data protection methods in biometric systems need to verify three other properties. First, the stored templates need to be easily and effectively cancelable if these become compromised (this property is called cancelability or revokability) [6], [7]. Additionally, the transformation from trait measurements to templates should be as close to irreversible as possible, as it should be impossible or infeasible for attackers to retrieve an approximation of the original trait using a compromised template (non-invertibility property) [4], [5].…”

Unimodal‐Bio‐GAN: Keyless biometric salting scheme based on generative adversarial network

“…There is a variation for the performance of the applied salting schemes. As shown by the figure, the recognition accuracy of the proposed Bio-GAN outperforms the Biohashing [5], Bloom Filter [7], Bioencoding [3], and Hetro_Convolved [19] schemes. While the best recognition accuracy of the applied salting schemes is achieved by randomized bit sampling [30] followed by Hetro_Xor-ed However, the authentication process of the schemes proposed in the Randomized bit sampling [30] and Hetro_Xor-ed [18] depends on an external token and a stored encrypted key, respectively, which make them vulnerable to the security threats of key-based schemes.…”

Secure Triplet Loss: Achieving Cancelability and Non-Linkability in End-to-End Deep Biometrics

“…Beyond accounting for natural biometric characteristic variability, biometric data protection methods need to verify template cancelability, non-invertibility, and non-linkability. Cancelability (or revokability) means the templates can be easily and effectively rendered useless if they become compromised, generally through the change of a personal key that is bound with the template [13], [14].…”