3. COMPUTER PROGRAM 
The computer program DIMIC-SH can handle up to a whole Landsat scene 
(ground area of about 185 km x 185 km). In view of the several constants, whose 
values are based on the Landsat system (see for example, Table 1), it assumes 
that the classification data are of the grey values range 0 - 255. The format 
of the original computer compatible tape (CCT) as obtained from the EROS Data 
Center, USA is not acceptable. For example, for a whole Landsat scene, each of 
the 2,340 rows must contain 13,056 grey values (that is 4 for each of the 
3,264 pixels), all four grey values of one pixel coming before those of the next 
pixel. (No "interlieving"). 
Fig. 6 shows the program flow chart, where NT1, NT4, NT5, NT6 and NT7 are 
storage files. NT6 and NT7 are the input and the output magnetic tapes respec- 
tively. Both have the same format, except that each record (or row) in NT6 is 
4 times longer than in NT7. This is of course because, before the classifica- 
tion each pixel is defined by 4 grey values, and after the classification by 
only one grey value. The grey value for each class in the result (NT7) had been 
specified on an input card by the user. 
  
The result of the classification, that is the magnetic tape NT7, is all 
that is required for the production of a film using a film recording equipment. 
This film serves as the negative in the production of a colour or black-and- 
white classification map. The negatives of the photographs presented later in 
this paper were produced on the Optronics P-1700 Photomation system at the 
University of Hannover, West Germany. 
3.1 Computations 
The top left hand corner and bottom right hand corner coordinates (in row 
and column numbers) of every training and test field are read in from cards 
and used by the subroutine CLASS for the generation of the coordinates of all 
training and test pixels which are then arranged according to classes and stored 
on file NT5. 
  
Using file NT5, the subroutine REALDA extracts from the magnetic tape NT6 
the grey values of all training pixels, and stores these on file NT1 according 
to classes. All data on NT1 are now preprocessed (see section 2.4.3) by the 
subroutine PREPRO and put back on this file as a replacement for the old data. 
All matrices computed and used for the preprocessing are saved, since they will 
be required later for the preprocessing of every pixel g on NT6 just before the 
computation of its decision values (eqn. 2.13) needed for its classification. 
The subroutine SEPHYP computes the coefficients of the polynomials (see 
section 2.2). Before this however, it calls on the subroutine DECIDE to deter- 
mine the degree of polynomials and the widths of the dead zones. NIS ic 
-cla 
The subroutine CLASFI assigns a class to every pixel contained on NT6 and tape 
stores the whole classification result on file NT4. It also compiles information 
on the number of pixels in each class, and on the total number of unclassifiable 
pixels (rejects). PEFORM reads through file NT5 containing the "ground truth" 
(that is all training and test pixels) and compares the class of every pixel 
with the computed class as contained on NT4, to produce the following statements 
which give an idea of the accuracy of the classification: e 7 
e 
- percentage of misclassified training-pixels for each class, ficat 
- misclassified test-pixels and rejected test-pixels as percentages of total of Ha 
number of test-pixels, 
- performance (or confusion) matrix for all test-pixels. 
64 
PPM.