Combination of Sources
Alain Appriou · 2014
The combination of information fragments drawn from different sources is a central and crucial function in data fusion, but also the focus of a number of difficulties. This chapter paints a clearer picture of the landscape of combinatorial rules existing in the different theoretical frameworks, to clarify their position relative to one another, specifying their practical properties and use, and put forward a general approach to combination which brings these frameworks together in the same formalism. Probabilities provide a fully integrated tool, directly exploiting measurements to determine the probabilities of the events being assessed. This tool, the Bayesian inference, is a rigorous approach, well suited to stochastic-type measurements. The connections between the possibility theory and fuzzy set theory enable us to transpose a number of the techniques. Zadeh's paradox highlights the paradoxical conclusions that can be arrived at from a situation of sharp conflict in the context of belief functions.