67 
3. Implementation 
Alexander Is written mainly in the local 
Archimedes “dialect" of the widely used Basic 
language. It offers most of the possibilities of 
languages such as Pascal while retaining the 
advantages of typical Basic features. One of 
the reasons for using Basic is that program 
ming the WIMP environment is not supported 
in Pascal or Fortran as much as it is in Basic 
or C. For reasons of speed, critical loops and 
procedures are implemented in Archimedes 
Assembly Language. Because of the RISC 
concept, this Assembly contains only a limited 
set of instructions and is thus comparatively 
easy to leam. However, some procedures are 
implemented in C and are easily integrated in 
the system. 
Version 1.0 of the package contains mainly 
image processing functions. Programs provide 
for simple radiometric corrections, feature 
extraction, image calculations, filtering and 
classification. The display program shows 
clearly the capabilities of the system. Colour 
composites are created within a few seconds. 
By changing the hardware palette, it is 
possible to find the best colour range to 
display the image. A selection can be made 
between 3-3-2, 3-2-3 and 2-3-3 for number of 
bits per pixel for the primary colours red, 
green and blue. The 3-3-2 selection, for 
example, yields a colour composite build up 
by 8 levels of intensity for red. 8 levels for 
green and 4 levels for blue. Greytone images 
are displayed by 16 levels for intensity. A 
pseudo colour scale uses 64 colours. Another 
special hardware palette setting is used for 
anaglyphic display of stereo images. The 
lefthand image displayed with 8 levels of 
intensity for red is mixed with the right hand 
image with 16 levels of intensity for green. 
This way the terrain relief can be perceived 
when using red and green filters (glasses) for 
channel separation. 
Though restricted to 8 bits per pixel, the 
display of an image (on multisync monitors) 
compares favourably with 24 or 32 bit display 
systems. Computation of a vegetation index, 
image calculations or linear and non-linear 
filtering are achieved within a few seconds. 
The local maximum within a 3 by 3 kernel of a 
256 by 256 image is done in 1 second. For 
version 1.0, all image arithmetic is performed 
in fixed point integer, computing with 3 
decimals. In and output data are raster files 
with one byte per pixel. Figure 5 shows the 
programs currently implemented. 
The package standardises on images of 256 
lines by 256 columns, though 512 by 512 is 
also supported, provided the user has at least 
2 Mbyte RAM. A multisync monitor is needed 
to display the full image, whereas, to view the 
entire image on the medium resolution moni 
tor, you would have to zoom the image or 
scroll the window. 
An image consists of a header file plus the 
respective raster files. The header files reside 
in directory INF, the raster files in directory 
RAS. The header file contains information like 
number of spectral bands available, number 
of lines and columns, data source, etc. 
Example: The image with the name AUS has 
4 spectral bands of Landsat TM which are 
bands 2. 3, 5, and 7. This is recorded in the 
header file. The respective raster files are 
then named AUS2, AUS3, AUS5, AUS7. An 
image is selected by its name, the bands are 
loaded by specifying the band numbers. The 
raster data structure is line sequential (similar 
to sprites, but without header) with 1 byte per 
pixel. The header file can be read with the 
program “View header". 
Version 2.0 will include more advanced image 
processing functions as well as GIS tech 
niques, e.g. m1m2l transformations, RGB to 
HSI, overlay of maps, digitization and distance 
analysis. Image arithmetic will be extended 
with binary and logic operations.