Bayesian Belief models for integrating match scores with liveness and quality measures in a fingerprint verification system

Recent research has sought to improve the resilience of fingerprint verification systems to spoof attacks by combining match scores with both liveness measures and image quality in a learning-based fusion framework. Designing such a fusion framework is challenging because quality and liveness measures can impact the match scores and, therefore, the influence of these variables on the match score has to be modelled. Further, these measures themselves are influenced by many latent factors, such as the fabrication material used to generate fake fingerprints. We advance the state-of-the-art by proposing two Bayesian Belief Network (BBN) models that can utilize these measures effectively, by appropriately modelling the relationship between quality, liveness measure and match scores with the consideration of latent variables. We demonstrate the efficacy of the proposed models on the LivDet 2011 fingerprint spoof dataset.