5.2
The labeling of a scene in terms of a model object is denoted by
where elements in
index image
features and those in
do model object features plus the NULL
. It
is also interpreted as a relational mapping from
to
( cf.
(5.5). We assume that f is a
realization of an MRF with respect to
. Below, we derive its
posterior probability using the MAP-MRF approach, in which contextual
constraints not only on the prior configuration but also the observation
are considered.