User's Guide

SPECSTRT

Stratifies a multispectral image based on the strata defined in a classified (categorized) image

Function:

Categorizes the pixels of an input multispectral image and produces a separate multispectral output image for each of the classes, each pixel's category being specified by a second input classified (pattern) image. Any data type (BYTE, INTEGER*2, INTEGER*4,or REAL*4) may be processed.

Parameters:

IN

Input image. The input multispectral image.

MASKIMG

Input mask (pattern) image. Name of the classified image, usually produced by BAYES, to be used along with the multispectral image to produce output multispectral class images.

OUT

Output image name prefix. The class names are appended to the output file name. For example, if the class names were WATER and LAND and the output file name is BALT, then output images BALT.WATER and BALT.LAND will be produced. The number of lines in the output classified images may be different depending on the number of pixels in the corresponding class. They will have the same number of samples per line as the pattern and input images. The last line may be partially zero filled.

CLASSNAM

Class name(s). At most, eight classes may be specified per run. The class name is appended to the output file name specified for the parameter OUT.

MASKVAL

Mask value(s). Grey levels to correspond with the given classes. Grey levels in the pattern image for the classes given in CLASSNAM. If a specified grey level is not found in the pattern image, the program continues processing with the next grey level.

Example:

1. LAS> specstrt in=balt.grp maskimg=balt.clas out=balt classnam=(water,land,forest,agri) maskval=(1,2,3,4)

   SPECSTRT uses the pattern image BALT.CLAS to extract the required classes from the input multispectral image BALT.GRP to produce output multispectral images BALT.WATER, BALT.LAND, BALT.FOREST, and BALT.AGRI.

Description/Algorithm:

SPECSTRT produces a separate multispectral output image for each class with pixel values as in the input multispectral image. It reads those parts of the pattern and multispectral input images corresponding to a class and transfers each multispectral pixel to the next sequential position in the appropriate output image of that class. Note that the spatial information of the input multispectral image is lost in creating the output image(s).

The number of output multispectral images is the number of classes requested, and each output image has the same number of bands as the input multispectral image.

The dimension of each band of the output image is NLi x NP, where NP is the number of pixels per line in the input image and NLi is the number of lines required to accommodate all the pixels in the corresponding class. Note that the last line may not be full and may have to be filled with zeros. SPECSTRT reports the number of zero fill pixels used in the last line of each output image.

Nonfatal Error Messages:

1. [specstrt-size] Pattern and input image sizes do not match... (NL, NS) = (XXX, YYY) used

   The smaller of the two sizes is used for processing.

2. [specstrt-noout] Specified grey level XXX not found in pattern image; no output image produced for class YYY

   Check the values specified for MASKVAL. Verify the input pattern image using LIST.

3. [specstrt-zeros] Output image will have XXX zeros on the last line

   The output image will have the same number of samples per line, NS, as the pattern and input images. If the number of pixels in a class is not an exact multiple of NS, the output image corresponding to that class will have a number of zero fill values in the last line.

Fatal Error Messages:

1. [specstrt-sameval] Same grey level values specified for different images

   Specify unique gray levels for each class.

2. [specstrt-patbands] Multi-band image not allowed for input pattern image

   Respecify single-band input pattern image.

3. [specstrt-maskval] Number of classes and number of grey values are not equal

   Respecify the CLASSNAM and MASKVAL parameters to contain the same number of values.

User Notes:

1. A typical processing scenario follows.

       To obtain broad (Level I) classification statistics
   
       o  Run ISOCLASS to obtain classification statistics.
   
       o  Run EDITSTAT to reduce and combine statistics to a smaller
          number of classes; e.g., the four classes WATER, LAND, 
          AGRI, and FOREST.
   
       o  Run BAYES to reclassify the entire area.
   
       o  Run SPECSTRT to produce a separate multispectral raw image
          for each of the classes; e.g., four multispectral images 
          are produced containing all the raw data in the original
          multispectral image.
   
       To obtain detailed (Level II) classification statistics, 
       repeat the following on each class image produced by 
       SPECSTRT.
   
       o  Run ISOCLASS on the image to obtain classification 
          statistics.
   
       o  Run EDITSTAT to reduce and combine statistics from 
          ISOCLASS to a smaller number of classes.
   
       o  Run BAYES to reclassify the entire area.
       o  Run SPECCOMB to recombine and map the Level I and all 
          Level II classifications into a single multilevel 
          classification.