Go to the USGS Home Page

USGS MIPS Classy Documentation






This program uses three input images  to create a classified image using an unsupervised clustering classificaion procedure. First, it collects histograms  of the three images and then uses the histograms to identify a min and max range for each image. A 3d histogram is then generated by compressing the min/max ranges into an NxNxN cube (N=25 to 50). Next, the program automatically searches through the 3d histogram to find modes/peaks in the 3d histogram; the modes become the center points for the various classes, so their positions are kept for use in the next step. The user specifies how many classes/modes are desired (NCLASS); the NCLASS modes selected are those with the largest number of pixels mapped to them. These modes are then used to classify the image.

The modes with the most number of pixels (the first NCLASS of them) are used to classify the input images in the following manner. All selected modes are assigned a unique class number, starting at the 3D histogram bin closest to (1,1,1). Bins directy adjacent to the mode are assigned to the class number of the given mode. Bins left unclassified are handled as follows: A 'route' is traced from the given bin to either a mode/class or until a classified bin is encountered using the maximum positive gradient. All bins along the route are set to the class of the first classified bin or mode encountered. A route may have a dead end (i.e., not lead to a class) in which case the bins along the route are set to the mode/class which is nearest (minimum distance) to the dead end bin.

  

Files

This program requires as input three unsigned 8-bit  image files and will generate one unsigned 8-bit image file. The following information is transfered from the first input file to the output file(if appropriate):

Image labels 
Mosaic labels
Projection labels
The history  from each of the input files is also transferred to the output file.

      

Parameters

MIN and MAX

These parameters are arrays used to select the minimum and maximum dn  values for each input file. The default automatic selection for the input files are calculated from their histograms by finding the dn closest to 0.5 percent for the min and 99.5 percent for the max. If dark or bright cover types are to be optimized, the user can set hist/her min/max parameters accordingly.

NCLASS

This parameter is used to select the number of classes the user wants in the output image. The default number of classes is 20.

SIZE and MSIZE

These parameters are used to select the 3d histogram and mode search dimensions. The histogram dimension is the actual size that the 3d histogram will have. The default histogram dimension is 50, which translates to a 50x50x50 cube. The mode search dimension is the size of window to look about for modes or peeks within the 3d histogram when calculating the required statistics for the mode search. The default mode search dimension window is 3 (i.e., 3x3x3).

Examples

From left to right, Original Landsat TM Bands 2, 4 and 5.

[Example Image] [Example Image] [Example Image]



Classified image result colorcoded for enhancement.

[Example Image]

See Also

Classavg 
Zip 
U.S. Department of the Interior | U.S. Geological Survey | Jobs
Home | Search | Disclaimers & Privacy Web Rings
Send comments to webman@TerraWeb.wr.usgs.gov
URL:
Last Modified: Thu Oct 31 15:08:21 MST 2002
FirstGov: Your First Click to the U.S. Gov ernment