Protection of Biometric Data from Presentation Attack

Abstract

An individual s unique physical and behavioural traits are measured and statistically analyzed in biometrics. Primarily, the technology is employed for the purposes of identification, access control, and monitoring subject identity. Traditional recognition techniques in security systems have been overtaken by biometric-based recognition. These days, a range of products use biometric technologies for various security purposes and incorporate more use of iris recognition (IR). Over the last ten years, there has been a notable improvement in the efficacy of infrared devices. However, there are security issues with the widespread use of biometrics, such as presentation attacks that attempt to limit system functionality. To avoid such a problem, state-of-the-art existing biometric types, attacks, datasets, and tools and techniques are analyzed. We design the detection model by proposing the deep learning technique that is Dual Channel Convolutional Neural Network (DC-CNN), which provides the generalization capability to detect the presentation attack and to enhance the model s accuracy were as outcomes of the detection model applied to the dataset sample show satisfactory results, with the True Detection Rate (TDR) being 98.70% on the LivDet-2015 publically available dataset. The prevention model is designed using Multi-Channel DenseNet (MC-DenseNet) techniques to prevent the presentation attack from entering the authorized device where the proposed method performs better than cutting-edge methods, 98.80% on LivDet-2015 and 99.16% on LivDet-2017 for true prevention rates, respectively. The Two-Channel Densenet (TC-DenseNet) is used for the detection of iris presentation attacks and prevention models, and it provides security to the system; experiments con- ducted on two popular, publically accessible datasets (LivDet-2017 and LivDet-2015) have validated the efficacy of the proposed method for iris PA identification. Outperforming the state-of-the-art techniques, the recommended method achieves a TDR of 99.16% on...