Toward Constructing Cancellable Templates using K-Nearest Neighbour Method

The privacy of biometric data needs to be protected. Cancellable biometrics is proposed as an effective mechanism of protecting biometric data. In this paper a novel scheme of constructing cancellable fingerprint minutiae template is proposed. Specifically, each real minutia point from an original template is mapped to a neighbouring fake minutia in a user-specific randomly generated synthetic template using the k-nearest neighbour method. The recognition template is constructed by collecting the neighbouring fake minutiae of the real minutiae. This scheme has two advantages: (1) An attacker needs to capture both the original template and the synthetic template in order to construct the recognition template; (2) A compromised recognition template can be cancelled easily by replacing the synthetic template. Single-neighboured experiments of self-matching, nonself-matching, and imposter matching are carried out on three databases: DB1B from FVC00, DB1B from FVC02, and DB1 from FVC04. Double-neighboured tests are also conducted for DB1B from FVC02. The results show that the constructed recognition templates can perform more accurately than the original templates and it is feasible to construct cancellable fingerprint templates with the proposed approach.