Selection of Experts for the Design of Multiple Biometric Systems

Designing multiple biometric systems: Measures of ensemble effectiveness

“…In addition, reported experiments will also show the extent to which the four considered combination methods are suited for combining multiple biometric experts. This paper extends the content presented in (Tronci et al, 2007c) by improving the analysis of the influence of the measures on the performance of the combined systems. In addition, we also improved the formulation of these measures and included experiments with an updated version of the DSS technique.…”

Dynamic score combination of binary experts

“…We considered ensemble made up of 3 multi-modal experts (i.e., different biometric traits and different algorithms). In the case of the NIST dataset we used all the possible combinations of experts, while in the case of the FVC2004 datasets we sorted all the possible ensembles according to the value of the d , that is a measure of the separation of the distributions of positive and negative score, the larger the d , the higher the separation [9]. Then, we selected the 10 pairs with the largest value of d , and the 10 pairs with the smallest value of d .…”