The objective of the feature subset selection step
is to determine the most inexpensive set of features
that allows one to separate an object of interest
from its neighbors with an error equal to or
smaller than a user-provided threshold e. This is a
two class problem (object, background) for which
prior information is available.