Multimodal Biometric Template Protection: Template Security and Recognition Performance

Biometric recognition is accepted as a reliable, robust and convenient means of person authentication and has been employed in many practical applications ranging from personal computer logons to large-scale border controls. The popularity of multibiometric systems, which make use of multiple biometric modalities information to reach a conclusion, has recently grown because they offer a better recognition performance than unibiometric systems. Several large-scale biometric systems such as that used at the Shenzhen-Lo Wu border control in China, the FBI-IAFIS in the USA and the UIDAI System in India have been installed. With the growing use of multiple biometric modalities, there is a rising concern about the security and privacy of biometrics data collected from the public. Because each person has a unique set of biometrics (e.g. fingerprints, face and iris), if these biometric data are compromised, it is difficult to revoke and re-issue new biometric data. The public acceptance of biometric technology not only relies on the accuracy and robustness of a biometric system, but also depends on the security of the stored biometric templates.

The central problem in this project is to ensure multimodal biometric templates (a biometric template refers to a set of features extracted from raw biometric data) are secure such that recognition performance is not affected. All existing methods offer a trade-off between template security and recognition performance. In the worst case scenario, the recognition performance of a multibiometric system with template protection is not as good as that of a unibiometric system, and therefore defeats the purpose of employing multiple biometrics. Unlike existing protection methods which provide only binary information on each biometric modality, the new multibiometric template protection method to be developed in this project envisaged here could provide a matching score for each biometric modality. Existing score-level fusion algorithms could therefore be applied to maximize recognition performance. The proposed method will be developed by taking advantage of transform-based and biometric cryptosystem approaches in biometric template protection, as well as traditional cryptographic algorithms.

The proposed project will culminate in development of a multimodal biometric template protection scheme and algorithm. Moreover, to address the limited number of publicly available multibiometric databases, this project will create a new multimodal biometric database available to the public. The results of this project and the new multibiometric database will contribute to security and privacy issues arising from and evaluation of multibiometric systems.