A New Multi-filter Framework with Statistical Dense SIFT Descriptor for Spoofing Detection in Fingerprint Authentication Systems

Fingerprint-based authentication systems represent what is most common in biometric authentication systems. Today's simplest tasks, such as unlocking functions on a personal cell phone, may require its owner's fingerprint. However, along with the advancement of this category of systems, have emerged fraud strategies that aim to guarantee undue access to illegitimate individuals. In this case, one of the most common frauds is that in which the impostor presents manufactured biometry, or spoofing, to the system, simulating the biometry of another user. In this work, we propose a new framework that makes two filtered versions of the fingerprint image in order to increase the amount of information that can be useful in the process of detecting fraud in fingerprint images. Besides, we propose a new texture descriptor based on the well-known dense Scale-Invariant Feature Transform (SIFT): the statistical dense SIFT, in which their descriptors are summarized using a set of signal processing functions. The proposed methodology is evaluated in benchmarks of two editions of LivDet competitions, assuming competitive results in comparison to techniques that configure the state of the art of the problem.