Either image classification methods are performed in two
steps. The first step in an unsupervised image
classification is performed by
i.cluster; the
first step in a supervised classification is executed by
the GRASS program
g.gui.iclass. In both cases,
the second step in the image classification procedure is
performed by i.maxlik.
In an unsupervised classification, the maximum-likelihood
classifier uses the cluster means and covariance matrices
from the i.cluster
signature file to determine to which category (spectral
class) each cell in the image has the highest probability
of belonging. In a supervised image classification, the
maximum-likelihood classifier uses the region means and
covariance matrices from the spectral signature file
generated by
i.class, based on regions
(groups of image pixels) chosen by the user, to determine
to which category each cell in the image has the highest
probability of belonging.
In either case, the raster map layer output by
i.maxlik is a classified image in which each cell
has been assigned to a spectral class (i.e., a category).
The spectral classes (categories) can be related to
specific land cover types on the ground.
The program will run non-interactively if the user
specifies the names of raster map layers, i.e., group and
subgroup names, seed signature file name, result
classification file name, and any combination of
non-required options in the command line, using the form
Alternatively, the user can simply type i.maxlik
in the command line without program arguments. In this case
the user will be prompted for the program parameter
settings; the program will run foreground.
This program runs interactively if the user types
i.maxlik only. If the user types i.maxlik
along with all required options, it will overwrite the
classified raster map without prompting if this map
existed.
g.gui.iclass,
i.cluster,
i.gensig,
i.group,
i.segment,
i.smap,
r.kappa
Last changed: $Date$
where each flag and options have the meanings stated below.
OPTIONS
Parameters:
NOTES
The maximum-likelihood classifier assumes that the spectral
signatures for each class (category) in each band file
are normally distributed (i.e., Gaussian in nature).
Algorithms, such as
i.cluster,
i.class,
or i.gensig,
however, can create signatures that are not valid
distributed (more likely with
i.class).
If this occurs,
i.maxlik
will reject them and display a warning message.
EXAMPLE
Completion of the unsupervised classification of
a LANDSAT subscene (VIZ, NIR, MIR channels) in North Carolina
(see i.cluster manual page for the first part):
i.maxlik group=my_lsat7_2002 subgroup=my_lsat7_2002 sigfile=sig_clust_lsat2002 \
class=lsat7_2002_clust_classes reject=lsat7_2002_clust_classes.rej
# Visually check result
d.mon wx0
d.rast.leg lsat7_2002_clust_classes
d.rast.leg lsat7_2002_clust_classes.rej
SEE ALSO
The GRASS 4
Image
Processing manual
AUTHORS
Michael Shapiro,
U.S.Army Construction Engineering
Research Laboratory
Tao Wen,
University of Illinois at Urbana-Champaign,
Illinois