Spiral Cube for Biometric Template Protection

In this paper we present a new approach for biometric template protection. Our objective is to build a preliminary non-invertible transformation approach, based on random projection, which meets the requirements of revocability, diversity, security and performance. We use the chaotic behavior of logistic map to build the projection vectors using a new technique that makes the construction of the projection matrix depend on the biometric template and its identity. Experimental results conducted on several face databases show the ability of our technique to preserve and increase the performance of protected systems. Moreover, we demonstrate that the security of our approach is sufficiently robust to possible attacks. 1 Introduction The growing concern for the problem of identity theft and the urgent need for individual privacy make the conception of personal authentication / identification systems increasingly important. These systems must authenticate users respecting several requirements, like speed, reliability, accurately and protection of user's privacy. Traditional systems of personal authentication which use passwords or ID cards are not able to meet all these requirements. For against, authentication systems based on bio-metrics, which use physiological (face, iris, etc.) and behavioral (signature, etc.) modalities , have proven a priority over traditional systems. But while biometrics ensure uniqueness, they do not provide the secrecy. For example, a person let his fingerprints on every touched surface and face images can be seen everywhere. Consequently, many attacks can be launched against the biometric systems, which reduce the credibility of these systems. Therefore, although biometric technologies have inherent advantages over traditional methods of personal authentication / identification, the problem of ensuring the security of biometric data is critical. In practice, opponents exploit the structure of biometric systems to launch their attacks. All biometric systems consist of four main modules (Fig. 1): the sensor module. The feature extraction module that selects the most significant characteristics in an image sent by the sensor and builds a biometric template test. The module of the da