Spoofed/Unintentional Fingerprint Detection Using Behavioral Biometric Features

Abstract

Fingerprints are common biometrics in smartphones as they are used for access to the device itself, or for authentication in applications. While fingerprints provide many benefits, they are vulnerable to spoofing attacks. This paper investigates countermeasures to spoofing attacks that use live fingerprints without consent either by force or by theft. We used behavioral biometrics to differentiate between intentional and forced fingerprint authorization attempts. Data was collected from several sensors and the most discriminating one was the accelerometer. A total of six data subsets, each with about 100 instances were collected, four for testing and two for calibration. A corresponding six tests were made on the subsets, in addition to one test on the combination of feature vectors from all sensors before and after using Correlation-based Feature Selection (CFS) to reduce the number of combined features. We used Naïve Bayes, Linear-Kernel and Cubic-Kernel Support Vector Machines (SVMs), and Deep Neural Network (DNN) classifiers. For the accelerometer-combined data, the classifiers scored 61%, 81%, 88% and 94%, respectively showing the DNN as the most powerful classifier, and for individual runs, performance was higher. The investigation was successful in differentiating between intentional and forced uses of fingerprint authentication systems.