Strategies for weighted combination of classifiers employing shared and distinct representations
Ali Hojjatoleslami, Josef Kittler · 2002
We present a theoretical framework for combining soft decision outputs of multiple experts employing mixed (some shared and some distinct) representations of patterns to be classified. Using this framework combination strategies are developed and their error sensitivity studied We show that nonuniform weighting combination strategies can be derived by taking into account the confidence of experts in their output.