A Random Walk-Based Cancelable Biometric Template Generation

Abstract

The usage of fingerprint biometric-based authentication systems raises serious security concerns if the generated reversible feature vector is stored directly. For this purpose, irreversible or cancelable templates are generated either through random projection methods or by performing non-invertible transformations (one-way functions) over the feature vector. In this paper, we have utilized the randomness of the random walk concept for generating a non-invertible fingerprint template. In a random walk, the future step of an object is independent of the past pattern, which makes it unpredictable. Here, we have used this property to generate multiple vectors by defining the number of objects and steps based on the size of the feature template. Diehard tests are performed to ensure the randomness of the individual vectors. After further processing, the projection matrix was created, which generates the cancelable fingerprint template. The performance of the proposed method was assessed by determining the equal error rate (EER) on FVC 2004 databases. The results were found to be satisfactory.