ABSTRACT
This chapter introduces generic person recognition systems that use classindependent quality measures for improved classification performance of single and multi-classifier multimodal systems. The proposed system, known as Qstack, uses the concept of classifier stacking. In this scheme, a classifier ensemble is used in which the first classifier layer is made of baseline mono-modal classifiers, while the second, stacked classifier operates on features composed of the normalized similarity scores and the relevant quality measures. This work also puts forward new solutions to the problem of probabilistic modeling based on Bayesian networks for base classifiers and fusion classifiers, which incorporate biometric data quality assessment and reliability measures embedded in biometric systems in order to improve classification performance.
