Syntax

   FACTORY_DEF <ReaderKeyword> RasterSegmenterFactory
   [FACTORY_NAME <factory name>]
   [INPUT FEATURE_TYPE <factory type>
      [<attribute name> <attribute value>]*
      [<feature function>]*]*
   [SEGMENTATION_METHOD (Watershed)]  //at the moment only watershed is supported
   [SHARPENING_FILTER_ENABLE (YES|NO)]
   [SHARPENING_FACTOR <conductance term>]
   [SHARPENING_ITERATIONS <number of iterations>]
   [GRADIENT_MAGNITUDE_FILTER_MODE (PRECISE|FAST)]
   [WATER_LEVEL_MIN <watershed low water level>]
   [WATER_LEVEL_MAX <watershed high level>]
   [OUTPUT SEGMENTED FEATURE_TYPE <feature type>
      [<attribute name> <attribute value>]*
      [<feature function>]*]*

Overview

   The RasterSegmenter accepts features with raster geometry. All other geometry types will be rejected. 
   
   Each input raster feature will be processed independently.
   
   The RasterSegmenter executes the user selected segmentation algorithm on the input raster geometry
   and outputs one raster for each segment of each input raster.
   
   Segmentation is the process of selecting arbitrarily sized groups of pixels from the input image 
   based on intensity differences in the input raster image pixels. 

   There are several algorithms that can be used for segmentation, each with its own options.

   Each output segment is a raster feature containing pixel values for the segmented area and nodata or
   zero values for the remainder depending on the options selected.


Bands, cell type and Palettes
   
   Rasters may have one band or multiple bands. The RasterSegmenter can handle up to 4 bands.

   The rasters cells may contain any type but the algorithms are designed to handle floating point
   cells (Real32 and Real64). A warning will be issued when cells are in a non floating point type. 
   
   Rasters may have no palette, one palette, or multiple palettes. For more information on raster 
   structure, see Rasters (IFMERaster). 
   
   The RasterSegmenter will resolve palettes on input rasters prior to
   processing.

Algorithms
   
   Watershed

      Watershed segmentation partitions the image by considering the image as a function of its
      coordinates, using the intensity of the raster cells.
      
      It usually starts from a Gradient Magnitude version of the raster, which enhances the borders
      of the image, and then runs the actual Watershed Filter to produce an intensity based
      segmentation of the image. 
      
      The watersheds algorithm does not produce a single image segmentation, but a hierarchy of 
      segmentations.
      
      Different segmentations can be selected from this hierarchy by adjusting the max water level,
      while some noise in empty areas can be filtered by raising the min water level.
      
      Before the raster is processed by the Gradient Magnitude Filter, and the Watershed Segmenter,
      an image sharpening (Anisotropic Diffusion) filter may be applied to the raster (recommended 
      for optimal results with most images) to remove noise while preserving the main features of 
      the raster.


Parameters

   SEGMENTATION_METHOD

      Select the segmentation algorithm.

   Watershed segmentation parameters:

      WATER_LEVEL_MAX

         Floating point percentage value between 0.0 and 100.0. Sets the relative dept of metaphorical
         flooding of the raster.
         
         This parameter can be used to control the level of segmentation of the image. 
         
         Its probably worth setting this parameter first as it tends to be the one that has the most 
         impact on the output.
         
         Usually greater values of this parameter lead to a less segmented raster. The values of interest
         tend to be lower than 40. The default, 20, usually results in a granular segmentation. 
         
         Consider increasing if the scale level if the output is too segmented.
         
         Required.


       WATER_LEVEL_MIN

         Floating point percentage value between 0.0 and 100.0. Sets the absolute minimum height values used by
         the Watershed algorithm. 
         
         It's used to remove shallower areas of the image that are less of interest. 

         Greater values set a higher minimum.

         A rule of thumb is to set this parameter to one undredth of the max water level and adjusting
         it from there.
         
         The default is set to 0.2 which should clear some of the noise without affecting the 
         prominent features of the image, but it depends heavily on the image, and the Output Scale
         Level.
         
         Increase to filter out more details. 
         
         Required.


      GRADIENT_MAGNITUDE_FILTER_MODE

         Control how the Gradient Magnitude filter is applied.

         Robust (Recommended): 
            (Principle Components), Use the largest two singular values of the Singular Value 
            Decomposition of the partial derivative.   
         
            This setting is slower but usually gives intuitively better results and it's the 
            default.

         Fast: 
            (Rooted Weighted Sum of Derivatives). Much faster and is parallelized (works well
            on bigger rasters), but not as robust. It may result in a noisy image.
         
            Try this if the other method is too slow.

         Gradient Magnitude Filter Mode is Required.



Output Tags

   The RasterSegmenterFactory supports the following output tags.

   SEGMENTED

      Output the segmented raster.

   REJECTED

      If there are any incorrect transformer parameters or the input is incorrect, the feature with
      that raster will be rejected