الفهرس 
Chapter 1 :IntroductionOnly 14 pages are availabe for public view
 |  | from | 130 |  |  |
| | | from | 130 | | |
Abstract
Biometric recognition is a difficult aspect of research, but it faces privacy and security issues. To solve this problem, cancelable biometrics have been proposed in the literature to make it difficult to obtain the original biometric images from deformed biometric images. Cancelable templates can be discarded or duplicated upon deletion. This thesis presents four algorithms for cancelable biometrics, and a proposed taxonomy were developed. We review various practical measures used for cancelable biometrics and provide mathematical formulas. This thesis presents four algorithms for cancelable biometrics.
1-The first and second algorithms depend on Viola-Jones algorithm for the segmentation of the region of interest in face images. This region is encrypted through chaotic scrambling and Advanced Encrypted Standard (AES) algorithm. This algorithm achieves high accuracy of recognition in a short time due to the reduction of the amount of computational time.
2-The second algorithm is a replica of the first one, but with feature fusion through deep learning. The deep learning is a powerful tool for feature fusion with non-linear fusion rules such as Multi-Exposure Fusion (MEF) algorithm. Hence, sophisticated patterns that represent both types of biometrics can be generated.
3-The third algorithm is a developed version of the first one. The difference is the inclusion of aliasing as a tool of intended deformation of data. The main merit of this algorithm is the low complexity at the cost of little deterioration in accuracy performance. This system is studied statistically even in the presence of noise. Simulation results reveal Equal Error Rate (EER) values down to 0.0028. In addition, Area Under ROC Curve (AROC) values up to 0.9945 have been achieved. The third proposed multi-biometric scheme has four stages: biometric acquisition and processing, Arnold’s Cat Map (ACM) encryption, decimation to reduce size, and finally interleaving of biometrics.
The acquisition and pre-processing stages aim at generating 2-D matrices for the four biometric traits of dimensions 128×128. The second stage depends on the ACM. The inherent characteristics of this map are used to perform pixel permutation. Pixel values are kept unchanged. The objective of this permutation process is to destroy the correlation patterns among pixels as a tool for hiding biometric patterns. The third step is the decimation process. This decimation process aims to destroy the Nqyuist sampling condition. The sampling in each encrypted pattern becomes beyond the Nqyuist rate. Hence, aliasing effect is introduced in the encrypted patterns. This effect can be considered as a tool for additional intended distortion of biometric patterns.
The last stage is the merging stage of four biometrics through an interleaving operation of rows and columns of the four biometric templates. These four versions are merged together to a single 3D matrix. Some sorts of decimation in rows and columns are applied to get two templates that have both encryption and aliasing effects. A sophisticated merging operation is implemented on these two versions to get a final 256×256 cancelable biometric template that can be used for biometric verification.
4-The fourth algorithm is a cancelable multi-biometric scheme that merges four biometrics for the same person into a single template. The basic idea of this scheme depends on the application of ACM on the four biometrics to achieve permutation. The obtained permuted versions are decimated by two to reduce the size of encrypted templates by four. Hence, the four encrypted templates are merged together through interleaving into a single cancelable template comprising the signature of each template. With this scheme, we ensure privacy of users, high security of recognition, low complexity, no redundancy in image size and small enrollment time. This interleaving process is implemented to keep the generated cancelable templates with the same dimensions as those of each of the original biometric templates.
Finally, the thesis presents a multi-biometric system that works on face, palm print, finger print and iris. Operation is implemented on each biometric, separately. In addition, dual finger print, dual iris, and dual palm print are considered. The region with maximum entropy is extracted from each image. For dual images, the pixels with maximum values are selected. After that, a concatenation process is performed. The resulting matrix after concatenation is encrypted with ACM through a permutation process. The last stage is the Double Random Phase Encoding (DRPE), which is based on Fourier transform and multiplication with phase masks. The main advantages of this approach are the utilization of multiple biometrics, in addition to the multiple levels of security. This system is studied statistically even in the presence of noise. Simulation results revealed EER values down to 0.00165. In addition, AROC values up to 0.991 have been achieved.