Cost-based feature selection for GIS-embedded data fusion

A powerful way of using existing information has been presented by Smits and Annoni (1999), who focus on a cost-based feature subset selection in the framework of the (interactive) verification and updating of land-cover maps with imagery from different space-borne sensors. Each feature comes with its own cost, and the user specifies a "budget" which the final feature vector may not exceed. As feature subset selection constitutes an important aspect of data fusion in general, this paper exploits further the concept of cost-based feature selection. The novelty is that the feature subset selection itself is also subject to a (computation) cost.