Design and implementation for secure embedded biometric authentication systems

Embedded personal authentication systems based on biometric information offers great security and convenience. The biometric processing is data intensive and the implementation on a resource-constrained mobile device is required to be efficient in terms of speed, memory, as well as energy consumption. Due to the uniqueness of the biometrics, the storage of the reference template is a key factor for the system security and especially for portable embedded devices. Thus it is essential to protect the template from possible attacks. Therefore the two major design challenges for developing secure embedded personal authentication systems based on biometrics is efficient implementation and system security.; The first part of this dissertation is about to efficient implementation for a fingerprint based authentication system targeting on embedded mobile devices. As the result of the HW/SW optimization, the execution time decreases from 9 second for fixed-point baseline algorithm to 4 second running on 50MHz LEON-2 microprocessor, gaining a 65% speed improvement. A memory analysis methodology is proposed to estimate the program run-time memory requirement, based on which various optimization techniques are employed, lowering the system memory requirement from over 1.5Mbytes to less than 500Kbytes.; The second part of this dissertation is dedicated to addressing the system security for embedded biometric authentication devices. A local structure based secure matching algorithm for fingerprint verification is introduced and the secure partitioning strategy is adopted to construct an oracle-based matching block using the logic level DPA-proof technique. Cryptographic biometrics are adopted to protect the biometric authentication system on the algorithm level. A template-protected fingerprint matching algorithm based fuzzy vault scheme is proposed, where a non-invertible form of the biometric features and a machine-generated random bit series is stored as the secret instead of the biometric templates. A micro-coded coprocessor, FV16, is designed to allow for a programmable yet high efficient implementation, which gaining 10-times cycle count improvement. In extension, the template-protected matching strategy based on the Error Correct Code (ECC) technique is proposed for iris recognition systems.