Cancelable knuckle template generation based on LBP-CNN

Security is a prime issue whenever biometric templates are stored in centralized databases. Templates are highly susceptible to varied security and privacy attacks. Unlike passwords, biometric traits are permanently unrecoverable if lost once. In this paper efforts have been made to generate cancelable knuckle print templates. To the best of our knowledge, this is the first attempt for generating secure template for this biometric-trait. Here for learning feature representation of a biometric sample, local binary pattern based CNN is used. The experimental results are evaluated on PolyU FKP knuckle database and demonstrate high performance. The proposed protected template is resilient to various privacy attacks as well as it satisfies one important criteria of cancelable biometrics i.e. revocability.