الفهرس 
AbstractOnly 14 pages are availabe for public view
 |  | from | 176 |  |  |
| | | from | 176 | | |
Abstract
Modern access technology nowadays is mainly directed to biometric authentication, rather than passwords or Personal Identification Numbers (PINs). The main advantage of dependence on biometrics in the authentication process is the ease of implementation as biometric traits always accompany the person without the need for mental or manual intervention. Biometric traits include signals such as voiceprint, Electrocardiography (ECG) and Electroencephalography (EEG); and images such as face images and fingerprints. Unfortunately, biometric authentication necessitates the storage of biometric traits or their features in databases. Hence, they become vulnerable to attacks. This thesis is mainly concerned with securing the biometric traits used in the authentication process.
Biometric security is accomplished through the utilization of biometric cryptosystems or cancelable biometric systems. Biometric cryptosystems can be used to hide information in biometric traits, while cancelable biometric systems can be used to generate deformed templates to be used instead of the original biometric traits. Both of them will be used in this thesis to preserve the privacy of users from hacking attempts. In addition, the authentication performance is maintained at high and acceptable levels. Both unimodal and multimodal biometric systems are presented in this thesis. Equal Error Rate (EER) and Area under the Receiver Operator characteristic Curve (AROC) are used to evaluate the systems performance.
The proposed systems in this thesis comprise cryptosystems, cancelable biometric systems, and hybrid systems. In addition, different encryption algorithms such as the Double Random Phase Encoding (DRPE), Optical Scanning Holography (OSH), and Deoxyribonucleic Acid (DNA) encryption are exploited in thesis. Each algorithm varies in terms of complexity, sensitivity to the initial conditions, and processing times. Encryption is utilized to save the biometric traits in an encrypted form as in cryptosystems or to aid in cancelable template generation as in cancelable biometric systems. Verification is performed through decrypted versions or cancelable templates.
Unimodal systems depend on a single biometric. Different algorithms such as Empirical Mode Decomposition (EMD) and 2-D conversion methods are considered in this thesis to generate the cancelable templates. The proposed systems can achieve a trade-off between the different cancelable biometric system requirements. DRPE and DNA produced different templates with different RPMs or initial values. Then, the diversity of templates can be achieved. Non-inevitability is achieved through signal spectrogram or polar plot in addition to the key-dependent encryption algorithms. DRPE allows a robust cancelable biometric system, and gives high verification performance, while EER is close to 0 and AROC is close to 1. Despite the robustness of the DNA encryption, it degrades the verification performance as the EER is degraded to 0.5.
Multimodal systems depend on combining two biometrics or more to generate the protected biometric templates. In this thesis, the Singular Value Decomposition (SVD) is the algorithm used to combine biometrics through data-hiding techniques. Data fusion, watermarking, and steganography are the different algorithms utilized to merge biometrics or their features. The proposed systems are tested in different domains. In data merging, the system performance in the wavelet domain is degraded with AROC values close to 0.7874 and 0.7988 in the cases of ECG and EEG, respectively. In watermarking, the proposed system has an accepted performance in different domains. However, the worst case occurs in the case of EEG signals in the Discrete Cosine Transform (DCT) domain with AROC equal to 0.9605. The proposed systems achieve a trade-off between the four different requirements of biometric systems. Finally, there are some differences between the proposed systems in complexity, processing time, and used biometrics. Application capabilities and requirements determine the appropriate system.