Selective fuzzy ensemble learner for cognitive detection of bio-identifiable modality spoofing in MCPS

Biometric features are widely used for user authentication and equally important to national and global technology systems. Various forms of biometric features, such as face, iris, fingerprint, are commonly used while more recently palm, vein and gait are also getting attention. Simultaneously various spoofing approaches have also been developed over time, which are capable of failing traditional biometric detection systems. Image synthesis with play-doh, gelatin, ecoflex etc. are some of the ways used in spoofing bio-identifiable property. Success of traditional detection systems are related to custom tailored solutions where feature engineering for each attack type must be developed. This is not a feasible process when we consider countless attack possibilities. Also, a slight change in the attack can cause the whole system to be redesigned and therefore becomes a limiting constraint. The recent success of machine learning inspires this paper to explore weak and strong learners with ensemble learning approaches using AdaBoost. Therefore, the paper proposes a selective ensemble fuzzy learner approach using Ada Boost, feature selection and combination of weak and strong learners to enhance the detection of bio-identifiable modality spoofing. Our proposal is verified on real dataset, LiveDet 2015, with a focus on fingerprint modality spoofing detection that can be used for authentication in Medical Cyber Physical Systems (MCPS).