Multibiometric systems: fusion strategies and template security

摘要: Multibiometric systems, which consolidate information from multiple biometric sources, are gaining popularity because they able to overcome limitations such as non-universality, noisy sensor data, large intra-user variations and susceptibility spoof attacks that commonly encountered in unibiometric systems. In this thesis, we address two critical issues the design of a multibiometric system, namely, fusion methodology template security. First, propose based on Neyman-Pearson theorem for combination match scores provided by matchers. The likeli-hood ratio (LR) test used directly maximizes genuine accept rate (GAR) at any desired false (FAR). densities impostor needed LR estimated using finite Gaussian mixture models. We also extend likelihood scheme incorporate quality samples. Further, show framework can be designing sequential systems constructing binary decision tree classifier marginal ratios individual achieves consistently high recognition rates across three different databases without need parameter tuning. For instance, WVU-Multimodal database, GAR rule is 85.3% FAR 0.001%, significantly higher than corresponding 66.7% best single modality (iris). use image further improves 90% 0.001%. Next, proposed applicable system operating identification mode. investigate rank level strategies hybrid utilizes both ranks perform scenario. While sources accuracy, it requires storage templates same user sources. Template security an important issue unlike passwords, stolen cannot revoked. Hence, securing entity fuzzy vault framework. have developed fully automatic implementations fingerprint-based secures minutiae iris cryptosystem iriscode templates. demonstrate better performance compared vault. example, our implementation fingerprint 98.2% less 0.01% provides approximately 49 bits values vaults 88% 78.8%, respectively. When stored separately, only 41 bits.