6.1
In this section, we assume that the data used for the parameter
estimation is due to only a single MRF, e.g.
a single texture. Or it is
due to more than one MRFs but has been classified into distinct subset
each of which is due to a single MRF. The data in the latter case could
corresponds to an image which has already been segmented. Such data is
considered as the known-class data. From the viewpoint of pattern
recognition, such estimation is supervised. The set of
parameters, 
, for each MRF, F, are estimated using the data
which is a clean realization (Some methods are needed to combine estimates obtained from more than one realization), f, of that MRF.
The algorithms described this section also provide the basis for more
complicated unsupervised estimation with unlabeled data.