Multimodal Template Protection Schemes for Iris and Fingerprint Biometrics

Abstract

Nowadays, with the advancement of technology, everything has been automated. Tra ditional authentication systems use ID cards, passwords, or PIN for identification. These systems have several limitations like password may easily be forgotten, hacked, or guessed, and ID cards could be misplaced, robbed, shared, or damaged. These limitations can be solved by using a biometric-based automatic recognition system. Biometrics is an area of science dealing with a person’s physical and behavioural characteristics. As these charac teristics are distinct for each user, they provide a reliable solution for authentication. The authentication systems which use a single biometric trait are known as unimodal biomet ric authentication systems. These systems face the challenges such as high security, poor recognition, and robustness against spoofing attacks. The systems which use more than one trait are known as multimodal biometrics authentication systems. These systems are more reliable, robust, and resistant to spoofing attacks. However, these authentication sys tems face challenges related to the privacy and security of the data. This biometric data is prone to various attacks like Hill Climbing, Brute Force, Record Multiplicity, Spoof ing etc. To safeguard this information, we use a technique known as Biometric Template Protection (BTP). This research uses Iris and Fingerprint to develop a multimodal authen tication system. These patterns are more resistant to genetic and environmental conditions throughout life. In addition, due to their randomness in pattern, there are fewer mismatches in the recognition system. This thesis presents a variety of multimodal template protection schemes designed for diverse applications that overcome various vulnerabilities and give a better balance of security and performance. The main objectives of this thesis include: (i) To develop a homomorphic encryp tion based template protection technique for high-security applications, (ii) To develop an alignment-free cancelable template protection technique, (iii) To develop a cancelable technique that uses deep CNN for enhanced security and performance, and (iv) To design a hybrid template protection technique that provides confidentiality and integrity. In this thesis, to achieve the mentioned objectives, we proposed a multimodal template protection technique using Local Random Projection and Homomorphic Encryption. Ini iv tially, features are extracted from both traits and combined to generate a fused template. Later, local random projection (LRP) is used on this template to create a reduced, revoca ble, and unlinkable template. Finally, fully homomorphic encryption (FHE) is applied to the created template to protect the user’s privacy, as all template operations are performed on encrypted data. Secondly, an alignment-free 3-D cancelable shell is developed for a multimodal authen tication system. First, features are extracted from the fingerprint. Then, using a user key set, a 2D spiral curve is generated from fingerprint features. Next, iris features are extracted using a pre-trained VGG-16 model and then random projection is applied to generate an iris feature vector. This generated feature vector is combined with the fingerprint shell to construct a secured 3-D shell. Thirdly, a cancelable template using Deep Binarization and Feature-Hashed Random Projection is developed for a multimodal authentication system. First, the features are extracted from both traits using a pre-trained Convolutional Neural Network(CNN) model. Next, these features are converted to binary codes by using deep binarization and combined on which AdditionRotationXor(ARX) operation is performed to generate fused features. The feature-hashed random projection is then used to generate secured templates. Lastly, a hybrid template protection techni que is developed using Block-Xor and sign cryption. Initially, features from both traits are extracted utilizing a pre-trained Convolu tional Neural Network (CNN) model. These extracted features undergo a deep binarization process, transforming them into binary codes. These binary codes are then fused using the Block-xor operation. To improve security, the signcryption operation is used to ensure the integrity and confidentiality of biometric data. The experimental results of this research demonstrate the effectiveness of template pro tection in biometric authentication system. The proposed methods outperform existing methods, offering improved security and performance. These developments can transform multimodal biometric template protection, offering better security and accuracy in authen tication across various applications, ensuring enhanced data protection and user privacy