![]() | Only 14 pages are availabe for public view |
Abstract Biometric security is new trend in models security systems. Biometrics in general comprise face, fingerprint, voice, palm print and other types. Traditional biometric security systems depend on feature extraction from biometrics, and then a feature matching strategy is applied in the authentication stage. Unfortunately, Traditional open biometric systems can be hacked, and original biometrics can be lost forever. That is why the trend of cancelable biometrics thus evolved as an alternative to open biometric systems. Cancelable biometric systems depend on the generation of encrypted or distorted versions of the biometrics to be used instead of the original biometrics in the verification process. In this thesis, we adapt to strategies to generate cancelable biometric templates. The first strategy depends on the utilization of the comb digital filters as a tool to distort the original biometric templates intentionally, while keeping the discrimination ability between biometrics. The comb filters are a multi-band filters that induces distortion in the original biometric templates. The comb filters orders are adapted until the best performance of the cancelable biometric system is achieved. On the other hand, the second strategy depends on generating distorted biometric versions according to the non-invertability of the aliasing effect. First, biometric patterns are decimated through a down-sampling stage. Hence, the sampling frequency is lowered beyond the Nyquist rate inducing the aliasing effect. After that interpolation is performed to maximize the amount of distortion in the biometric pattern. Polynomial techniques such as bilinear bicubic, cubic spline and cubic O-MOMS are used in the interpolation process. This strategy guarantees intended distortion of biometric patterns, while keeping discrimination ability. The common thread between both proposed strategies is the utilization of digital filtering in both of them. In the first strategy the comb digital filter with order adaption is used. On the other hand, the second strategy with Cubic spline and Cubic O-MOMS interpolation is implemented with digital causal and non-causal filter for digital interpolation coefficient estimation. Hence, the digital filtering is the heart of both proposed strategies. Simulation experiments are introduced in this thesis on both face and fingerprint images. |