Risk Inference Models for Security Applications

Abstract

This paper focuses on the causal graph models for machine reasoning and its applications to risk assessment in biometrics. Specifically, we consider probabilistic inference performed on video data, images, speech and other human biometric data. In our approach, called the Multi-metric Inference Engine, the Bayesian network are constructed using different metrics of uncertainty, such as point probability, interval probability, fuzzy probability, and Dempster-Shafer model. We demonstrate the Inference Engine techniques using biometric-enabled security scenarios and propose a software tool for experimental study.