الفهرس 
CHAPTER 1: INTRODUCTIONOnly 14 pages are availabe for public view
 |  | from | 161 |  |  |
| | | from | 161 | | |
Abstract
Speech or voice is a biometric candidate that can be used to recognize persons from
their spoken words. The main target of Speaker Identification (SI) systems is to distinguish
the person from his spoken words. Generally, the security of biometric systems is very
important in present-day life. Unprotected biometric templates raise concerns about
security and privacy, as for instance, susceptibility to attacks, cumbersome renewability,
and cross-matching. So, if these templates are attacked, the biometrics are lost forever.
Hence, there is a need to secure original biometrics by keeping them away from utilization
in biometric databases. Cancelable biometric schemes provide solutions to achieve privacy
preservation in the biometric recognition process. Cancelable biometric systems depend on
the transformation of data extracted from original biometrics into a new format, so that
users can replace their biometric templates in the same or different systems. This thesis
addresses the problem of cancelable SI. It introduces six cancelable SI systems based on
spectrogram estimation and different encryption schemes.
The first system depends on chaotic Baker map encryption, as it is a permutation
tool, which performs the randomization of a square matrix of dimensions N × N by
changing the pixel positions based on a secret key, sample unwrapping of elements in subblocks
to rows, and then final randomization of the array. The second system depends on
Radon transform, which is applied as a second step on the spectrogram of the signal
encrypted using Baker map. The Radon transform gives the new templates to be stored in
the database.
The third system relies on Rivest-Shamir-Adleman (RSA) algorithm for encryption.
The RSA based on the principle of multiple keys. The public key is utilized for message
encryption, and it can be visible to anyone, while the private key is utilized for message
decryption, and it must be kept confidential. Three stages are involved in RSA operation:
key generation, encryption and decryption.
Abstract
iv
The fourth system depends on the use of the Radon transform as an additional step
after the encryption phase with the aim of increasing the security of the system and making
the task difficult for any attacker. After using the RSA technology to encrypt voice signals,
we estimate the spectrograms of these encrypted signals, and then apply the Radon
transform to the spectrograms representing the new templates that will be stored in the
database.
The fifth system depends on the 3D chaotic map as a hybrid cryptosystem that
combines five stages to complete the overall encryption process for building a reliable and
robust cancelable biometric recognition system. The proposed five stages are 3D chaos
generation, chaos histogram equalization, row rotation, column rotation, and XOR
operation.
It is evident from the obtained results that the proposed systems are secure, reliable
and practical. They have good encryption and ability to generate cancelable templates.
These characteristics lead to good performance. The proposed cancelable speaker
identification systems are evaluated under the influence of Additive White Gaussian Noise
(AWGN) with different strengths. This makes them more accurate in identifying the users
and also more resistant to attack attempts. In addition, security is enhanced through
maintaining the confidentiality of the processed data.
The last proposed system relies on deep learning technology for cancelable SI. This
system introduces three deep pre-trained learning models based on Convolutional Neural
Network (CNN) for cancelable SI from spectrograms: AlexNet model, VGG16 model and
VGG19 model. The spectrograms are created for speech signals encrypted with Baker
map. We treat spectrograms like images, attempting to extract deep characteristics that
may be utilized as cancelable templates. Because of the large database, the suggested
system uses neural classification rather than threshold-based classification.
The introduced cancelable SI systems are evaluated based on probability density
function of genuine and imposter distributions. In addition, the Area Under Curve (AUC)
is used as a metric for quality of authentication. Simulation tests prove the high quality and
security of the proposed cancelable SI systems.