Classical image classi®cation has practical value
in remote-sensing applications only when applied
to very speci®c and rather simple (i.e., low number
of classes) image analysis problems. To-date, an
important part of the remote-sensing satellite image
interpretation work for land-cover database
generation is done manually by a photointerpreter
who heavily relies on additional data.
However, image analysis and pattern recognition
tools can become of potential bene®t regarding
the veri®cation or updating of existing landcover
maps, as often much prior information is
available. In this article, a method is proposed that
appropriately combines object characteristics and
prior information related to class labels and type
of sensor. A subset of features is selected for the
local image, based on explicit measures of accuracy
and computational cost. Experimental results
showed that the method is a useful tool for the
veri®cation and the updating of existing landcover
maps with image data.
For further reading, see (Duin, 1996; Lillesand
and Kiefer, 1987).