الفهرس 
AbstractOnly 14 pages are availabe for public view
 |  | from | 109 |  |  |
| | | from | 109 | | |
Abstract
Biometrics has been widely adopted in various applications that require identification or verification. Biometric system storage is the common attack target for impostors. To address this issue, several biometric protection systems have been proposed to ensure the proper use of biometric templates by protecting them against the misuse by attackers. Many of these systems require the biometric templates to be presented in a binary form. Unfortunately, most of biometric data are usually represented as real valued templates (infinite space) such as: face, hand geometry, and fingerprints. Therefore, extracting binary templates from real-valued biometric data is a key and fundamental step in biometric data protection systems. A new trend of researches in biometrics entitled ”Biometric Binarization” has been presented. The purpose of biometric binarization schemes is extracting binary representation from the real valued biometric templates to be used by the target biometric protection system. Also, the binarization process produces data representation that usually takes less storage capacity as well as reduced time for matching templates. Mapping from the continuous real-value domain to the discrete binary domain is non-trivial, especially when optimal performance and balance between security and discrimination are demanded which is the major challenge of such biometric binarization schemes. Biometric binarization is generally categorized as local schemes and global schemes. Local binarization scheme generate the binary templates through three stages. First, quantizing the original feature domain for each independent dimension into intervals. Then, assigning a binary code to each interval. Finally, the output binary string is generated by concatenating the binary code for each feature dimension. Because preserving the discrimination power of the generated binary string is important, both quantization and encoding should provide optimal performance when using HD classifier for comparing the resulting binary strings. On the other hand, global binarization scheme binarize the original real-valued biometric template using a series of transformation functions applied to the entire template. The main challenge for any global binarization scheme is how to construct these transformation functions while preserving or even enhancing the discriminability of the transformed binary templates.
In this work, the biometric binarization is considered as an optimization problem. We attempts to handle the existing challenges for biometric binarization by introducing two novels biometric binarization schemes based on optimization strategy. The first one is local-based biometric binarization scheme, the adopted optimization strategy aim to find the optimum combination for quantization levels and encoding functions for each feature dimension in order to optimally transform the original biometric data into binary representation. Whiles, the second proposed scheme is global-based scheme, it utilize an optimization strategy that search for the optimum coefficients for discriminant linear functions which used to binarize the original biometric data. The proposed schemes ensure the security and privacy requirements of biometric protection systems. Moreover, results obtained from the experiments conducted on face and fingerprint biometric databases show promising recognition accuracy.