ng 
ull 
he 
nd 
ng 
the 
ase 
ks 
ige 
ent 
1a 
ray 
the 
ble 
;'ed 
ict, 
ter 
ent 
the 
nay 
ger 
ach 
the 
bits 
led 
Jers 
wn 
na 
d to 
ged 
ume 
1g a 
the 
an 
tion 
dsat 
inal 
in 8 
  
  
bits per pixel. The region is rectangular with an extent of 
approximately 6100 by 5300 pixels per scene. If the scene 
has been geometrically corrected and geocoded, the map 
coordinates of each pixel can be determined by its position in 
the raster. 
An example of a layer is a collection of several regions, 
where each region belongs to the same channel of a different 
TM scene, and where each TM scene covers a different area 
of the ground. In effect, this layer would be a single- 
channel mosaic of TM scenes. An example of an image is a 
collection of such layers, where each layer belongs to a 
different TM channel. 
The image described above could be extended to include an 
additional layer consisting of, for example, regions derived 
from SPOT panchromatic data. Note that the raster 
definitions of the SPOT-derived regions would not be the 
same as those of the TM-derived regions (the SPOT-derived 
regions would have 10 metre by 10 metre pixels and their 
grids might not correspond to a regular subdivision of any of 
the TM-derived region grids). 
The layer of SPOT-derived regions might be extended to 
include regions consisting of digitized panchromatic aerial 
photographs where each pixel has a ground dimension of 
say 1 metre by 1 metre. Regions that belong to the same 
layer do not have to conform to the same raster definition 
any more than do regions that belong to different layers. 
Thus a layer might be composed of data from a variety of 
sources in order to assemble the most complete ground 
coverage possible for an area of interest. 
Regions may belong to more than one layer of the same 
image or even to layers belonging to different images. In 
any case, only one copy of a region is physically stored. 
Thus, images and layers are constructs that allow different 
relationships between regions to be defined. 
The descriptive data for regions and layers are stored in 
tabular form in ASCII files, and the actual image pixel values 
are stored in separate binary files as unstructured stings of 
bytes. The region table includes all data needed to impart a 
structure to the stored image pixel values, as well as the 
parameters of each regions' raster definition. This 
implementation allows all of the descriptive data to be read 
and modified using a text editor. 
3.2 Image Display Function 
In the current implementation, the raster image display 
function resides in the Cartographic Processing Package. 
Either a monochromatic or a colour composite display can be 
created by specifying a single or three image layers. An 
image layer specified may conceptually be composed of one 
or many regions, each of which is a physically separate 
collection of data. The display system is programmed to 
unify separate regions into a single layer when the display is 
created. 
For any given map display on a window-based display 
system, the drawing surface (or canvas) represents a certain 
geographic extent, depending on the scale at which the data 
is drawn on the screen and the geographic location that the 
data represents. Furthermore, each pixel in the canvas 
represents a certain geographic area and location. Thus, the 
canvas may be thought of as having a raster definition for 
which the pixel size, raster dimension, and location 
parameters are set by the display scale and location. When 
displaying raster-imagery as a backdrop to a digital map, 
CARIS obtains the canvas raster definition parameters as set 
by the map display and automatically resamples (by nearest 
neighbour) the image-region data to that raster definition 
when transferring the data from disk to RAM (Figure 3). 
Because of this resampling, the dimensions of the canvas 
777 
only influence the spatial resolution of the raster image 
display, and not the extent of an image-region that can be 
displayed at any one time. Also, for each image-region in 
the display, only the part that intersects the geographic extent 
of the display is read from disk. When reading a layer that is 
composed of multiple overlapping regions from disk, 
CARIS selects one region to be visible in the overlap area. 
The image display is controlled by the following commands: 
select layer(s), raster image draw, raster image on, raster 
image off and dim intensity. 
4. CARIS RASTER IMAGE EXTENSION 
Digital image processing and analysis operations were 
incorporated at three locations in the CARIS software as 
shown in Figure 3 [Derenyi, 1991]. 
4.1 Primary Graphics Raster Image Extension 
This enhancement to the Primary Graphics module CARED 
allows the display of geographically registered images as a 
backdrop to digital cartographic data. Digitization in the 
raster image backdrop and interactive cartographic editing 
can be performed. The full pre-existing functionality of 
CARED is preserved, so that operators who are already 
familiar with the system, do not have to relearn any aspects 
of its operation to use the extension. 
A number of real-time image enhancement routines were also 
implemented in CARED to facilitate the interpretability of 
images. These include contrast enhancement by histogram 
equalization, piecewise linear stretch, neighbourhood 
averaging, bi-directional gradient, and by user specified 
histogram. Pseudo-colour enhancement and transformation 
from red/green/blue to intensity/hue/saturation colour space 
are also available. 
4.2 Database Analysis Raster Image Extension 
This extension is attached to CARMAN. It facilitates the use 
of geographically registered image data as a source of 
information in interactive spatial analysis. Current 
implementation includes the generation of polygon based 
image statistics, supervised image classification, statistical 
evaluation of classification results and colour coded display 
of the classes. The image statistics provided are the mean 
vector, the variance covariance matrix, and the lower and 
upper limits of radiance values in each data layer. The 
statistics are generated for predefined areas represented by 
zones. À zone is a raster object generated in CARMAN or 
defined by boundaries stored in a polygon file. 
4.3 CARIS Raster Image System (CRIS) 
The operations described in the previous two subsections are 
executed in RAM, while a digital map and image appear in 
registration on the monitor. They are intended for 
applications which require visual interaction by the operator. 
Many image processing operations are, however, performed 
on images alone without using any companion cartographic 
data and without the need for interactive viewing. These are 
preprocessing tasks which prepare large data sets for the 
subsequent cartographic mapping and database analyses. 
Therefore, these operations can conveniently be executed on 
disk in batch mode and the results transferred to CARED or 
CARMAN for interactive manipulation. The new software 
module CRIS serves this need. 
Currently available raster image processing functions are: 
various contrast enhancement routines, including direct 
histogram specification; spatial filtering, including a number 
of post-processing options; arithmetic operations performed 
on any number of layers in the raster image file; application 
of a user-defined threshold to the raster image; principal 
component transformation and image processing in the