The input data are in MSS format; i.e. the bands are appended together in the sample direction using the program MSS. The pixels can be in any format since they are converted to real in the program. The MSS parameter specifies how many bands are in the input file.
The first output file is a statistics dataset (like those produced by
USTATS, etc) which contains the number of data points, the mean vector,
and the covariance matrix for each cluster.
The second output file is intended for the RESTART option. It contains
the input data for those bands that were used in the clustering and another
band which gives the cluster number that the pixel was assigned to. This
file is in MSS format and has the same pixel format as the input file. If
SIZE and/or INCR was specified on the first run, then the data will have
been compressed accordingly and these parameters must not be respecified on
a restart.
Standard VICAR size field
Starting line
Starting sample. This is PER BAND, i.e., it does not apply (as is standard for the MIPL SS parameter) to the entire input image, but (because the image is in this case in MSS format) to each band separately.
Number of lines
Number of samples. This is PER BAND, i.e., it does not apply (as is standard for the MIPL NS parameter) to the entire input image, but (because the image is in this case in MSS format) to each band separately.
The MSS parameter specifies how many bands are in the input MSS format file.
The BANDS parameter specifies which bands in the input file will be used for the clustering. This allows the user to try clustering various combinations of bands without having to make up a new input dataset each time. There must be at least two, but no more than twelve bands. Default is to use all bands in the input file in the order in which they occur in the file. However, if RESTART has been specified, then the last band is not used. I.e., default is: BANDS = (1,2,...,MSS) if not RESTART, BANDS = (1,2,...,MSS-1) if RESTART.
Increment in line and sample direction. This parameter specifies both LINC and SINC, q.v.
Increment in the line direction. This parameter specifies that only every LINCth line (starting with SL) will be used to select objects from the image for the algorithm. Default = 1 (every line).
Increment in the sample direction. This parameter specifies that only every SINCth sample (starting with SS) will be used to select objects from the image for the algorithm. Default = 1 (every sample).
THRESH specifies the minimum DN value that is to be accepted as a valid pixel. If an object has a DN less than THRESH in any band (of those specified by the BAND parameter), then it is rejected. The default is to not do thresholding; all pixels in size region are used. In the second output file, all pixels on a given line that have been rejected are written at the end of that line, with 0 DN. Hence THRESH does not need to be specified on a RESTART, even if it was specified on the first run.
The NCLUS parameter guides the algorithm in choosing the number of clusters (see theory of operation). The first number is the desired number of clusters and the second number is about how many clusters the algorithm can stray from the desired number.
The NCYCLE parameter is the maximum number of temperature cycles the algorithm will perform. If it is large then the algorithm will lower the temperature very slowly, but if it is too small the system will be quenched and will not be at a true minimum energy configuration. The default will usually be adequate.
The NITER parameter specifies how many iterations are to be done for each cooling cycle. The default is to do 4 times the number of data points, so that on average each point will be considered about four times per cycle. The default will usually be adequate.
The TEMP parameter gives the initial temperature to start the system off with. The initial temperature should be of the order of unity. A higher starting temperature will make the system behave more randomly at first. The default will usually be adequate.
To restart the run, the RESTART keyword is specified and the second output file from the previous run is used for the input file. Note that the bands in the new input file are in different positions than the bands in the old input file and the input includes an additional band containing cluster numbers, hence MSS will in general be different than on the first run. If SIZE or INCR (or equivalent parameters) were specified in the first run they must not be specified on a RESTART, as the data will have been compressed accordingly.