ABSTRACT

While multimodal biometric systems are considered to be more robust than unimodal ones but traditional fusion rules are more sensitive to spoofing attempts. The proposed system is designed to overcome spoofing in worst-case scenario where impostor was able to create fake biometric traits of both face and fingerprint modalities in the presented multimodal biometric system. The paper investigates median filtering fusion rule as a spoofing resistant alternative to traditional sum rule based fusion rules. Experiments on the latest face video database (CASIA Face Anti-Spoofing Database) and fingerprint spoofing database (Fingerprint Liveness Detection Competition 2015) illustrate that the given system is more robust to spoofing attacks than the existing anti-spoofing methods even when m out of n samples of both the biometric traits to be combined are attacked.