
SPECTRUM(HELP)               Khoros                SPECTRUM(HELP)

     OVERVIEW

     Spectrum is an interactive signal/image classification sys-
     tem.  It aids the user in classifying image data.  An image
     may be displayed, and then pixels in the image assigned to
     class categories by clicking on the desired pixel in the
     image. For identification, class categories are given a name
     and an unique color. A class category may contain as many
     pixels as desired; pixels may be moved from one category to
     another as desired.  When the classification procedure is
     complete, the image can be written out along with a legend
     file for later use.

     A "Print Class" capability gives information about class
     categories according to the pixel over which the cursor is
     positioned.  The "Legend Control" feature is used to modify
     the legend of class categories;  among other things, you may
     delete a class, empty a class, transfer an entire class into
     another one, create a "catch-all" category of remaining
     unclassified clusters, and hide selected class categories
     from view in order to examine the image without their
     interference.  A "Zoom" capability allows you to zoom in on
     the image, and select pixels to be assigned to a class
     category by clicking on the desired pixel in the zoom window
     rather than in the image.

     Spectrum was specifically designed for images that have been
     pre-clustered, and therefore have additional information
     associated with them, this info being stored in the map
     data.  This special info in the map data will include N rows
     of cluster centers (mean vectors), where each column of the
     info is associated with a special feature of interest that
     was extracted during the previously-used clustering algo-
     rithm.  If you do NOT have pre-clustered data, the two para-
     graphs above apply to VIFF images of data type BYTE, without
     maps of type FORCED;  if you do have pre-clustered images,
     there are additional features of spectrum that you may take
     advantage of:

     o display of different map columns as Red, Green, or Blue
     o display of functions of map columns as Red, Green, or Blue
     o a 2D scatter plot where one map column provides the X
     coordinates, and another map column provides the Y coordi-
     nates;  unassigned cluster numbers appear in white, while
     assigned cluster numbers appear in the same color as was
     specified for their class category in the legend
     o the same 2D scatter plot as above, but with specified
     functions of map columns providing the X and Y coordinates

     Please see the section on I/O if you have additional infor-
     mation associated with the input image provided by a clus-
     tering algorithm;  this section details the input format
     expected by spectrum for such information.  In addition,
     this section also details the ascii Legend File that is used
     with Spectrum.

     IMPORTANT:  the use of spectrum will vary according to the
     user's intent:  some will take advantage of the full capa-
     bilities of this program by providing additional information
     about a pre-clustered input image in its map data, as speci-
     fied in the section on IO;  others will use "normal" VIFF
     images of type byte that have not been pre-clustered, and
     therefore have no additional information stored in the map
     data.  When referring to a displayed element of the image,
     the correct term in the former case would be "cluster
     number", while the correct term in the second case is
     "pixel".  Because of unpredicatability of the audience,
     these terms have been used (somewhat incorrectly) inter-
     changably in the documentation.  To avoid confusion while
     reading documentation on spectrum, simply substitute "clus-
     ter number" for "pixel" if you are using spectrum for pre-
     clustered images with additional information stored in the
     map data;  substitute "pixel" for "cluster number"  when
     using "normal" VIFF images with no additional information.

     The graphical user interface of spectrum contains several
     subforms through which you may access the different features
     of spectrum; these subforms may be accessed by clicking on
     the desired subform button that appear on the left hand side
     of the graphical user interface.

     Subforms

          Input/Output        Used to input and output images and
                              legend files.  Can also be used to
                              output the colormap as it appears
                              in the image.

          Display             Used to change how map columns are
                              displayed when additional informa-
                              tion from a pre-clustering algo-
                              rithm is provided.

          Print Class         Gives information about class
                              categories according to the pixel
                              (cluster) over which the cursor is
                              positioned.

          Legend Control      Used to modify the legend of class
                              categories.

          Zoom On Image       Enables the user to zoom in on the
                              image.

          Scatter Plot        An alternative way to examine clus-
                              tering data.

          For more detailed information on each of these sub-
          forms, see the online help that is pertinent to the
          subform in question.

     RESTRICTIONS

     As this is the "first iteration" of the interactive classif-
     ication program, spectrum still has a few limitations. The
     following list describes the known limitations with spec-
     trum:

     1)   You must have image data of type BYTE.  Spectrum will
          not work with images of any other data type.

     2)   The maximum number of rows in the map data provided (if
          any) is 256.  If more than 256 rows are provided, they
          will be automatically reduced to 256 without prompting
          or warning.  You may have as many columns in the data
          as you like.

     3)   Spectrum will work best if the map data provided (if
          any) has values that are less than or equal to 255.
          That is, map data must be of type FLOAT, but its values
          should be within BYTE range.

     3)   To move the zoom window on the scatter plot, you must
          click exactly on one of the points in the plot.

     4)   Clustering routines provided with Khoros such as viso2,
          kmeans, and others are NOT compatible with spectrum at
          this time.  We are hoping to convert them soon, but
          cannot guarantee that they will be converted for Patch
          3 of Khoros.  Thus, if using any of these Khoros pro-
          grams to cluster your data, you will have to create a
          little intermediate program to convert the data to the
          format expected by spectrum (see the online help on I/O
          for more details).

          Copyright 1991, University of New Mexico. All rights
          reserved.

Khoros                    Release: 1.0                          1

