Ltized is the 
lat has been 
ssible to 
re the 
?hotos should 
led in a 
and displayed 
tep is to 
the CRT lines 
iter those 
ides on the 
displayed on 
the centro!ds 
tizer together 
or automatic 
software will 
area with the 
red map on the 
terpreter. 
tig areas and 
ecting the 
ing it, or by 
y be sent to 
g the 
e considered 
’s electronic 
g is 512 x 512 
CCD’s 2048 x 
ced in 1986 
x 3500 , but 
acy as yet. 
Is and, like 
wfinder for 
ems have the 
summarizes 
BIT fair 1986 
color OCR 
RGB 
RGB 
haracter 
on 
a SPATIAL 
deo digitizer 
esolution of 
0.3 mm 
y lower than 
.1 to 0.025 
mera has been 
otodiode array 
) that moves 
tween the 
each 
photodiode is converted into a digital value having 
up to 12 bits and the information is passed to the 
computer, as parallel data (DWYER, 1985). Solid 
state array sensors have excellent geometric 
properties that are a direct consequence of the 
photolithographic method of fabrication. (NAGY, 
1981). Eikonix claims a spatial precision of 1 
pixel, corner to corner. This has not yet been 
verified, but ITC plans to do a full calibration of 
its digital camera in the near future. The system at 
ITC includes further a normalizer board to 
compensate for individual diode characteristics and 
a computer driven color filter wheel enabling color 
separation. 
(Another Eikonix model uses a 4096 linear CCD array. 
The CCD however catches less greylevels and is less 
sensitive to color. A similar CCD camera is 
manufactured by Datacopy and was used by Chrisman in 
his earlier cited project.) 
Using the 2048 pixel camera on a 9 inch airphoto 
overlay means scanning with a resolution of 9 pixel 
per mm or about 0.1 mm which is acceptable for a 0.3 
mm line width and compares favourably with the 
effective accuracy of usual manual digitizing. 
5.3 The current pilot system 
Fig 3 gives an overview of the system as it is 
currently applied. The total system configuration 
used for ADIOS consists of the Eikonix camera, a PDP 
11/24 and an AED 767 graphics terminal with 8 
bitplanes. 
For the experiments photointerpretation overlays are 
used (fig. 1) that were made by H.T.J. Lutchman in 
1980 for a landuse change study (DE BRUIJN, 1981). 
They are typical for the output of such studies and 
were indeed made before there was any question of 
automatic digitizing. Hence it is expected that 
results obtained with those overlays will be 
representative for routine operational situations. 
The various steps of the procedure are as follows: 
Step 1 scanning 
In scanning the overlays a red filter has been used 
to eliminate some red wax pencil lines also present 
on the overlays but not relevant for the landuse 
interpretion. The filter reduces effectively the 
values of the red lines but they do not disappear 
completely and return to a certain extent after the 
edge enhancement in step 2. 
Step 2 edge enhancement and thresholding 
Although the scanned image looks quite good when 
displayed on the screen the numerical values of the 
pixels are affected considerably by 
- unequal lighting 
- varying density of overlay paper 
- varying line intensity and width 
In the present provisional setup a simple amateur 
photographers reprostand lighting system with 4 
incandescent 60 watt bulbs is used and it has been 
found that lighting varies indeed considerably. 
Some of the darker white areas have values below 
those of some of the black lines in the lighter 
areas on other scanlines. The values in the 
unprocessed scanfile vary approximately from 50 
(dark lines) to 200 (light paper). When simple 
thresholding is applied the lines loose their 
connectivity or the white areas will get a lot of 
noise, (fig.4) 
To "reconstruct" the linework a 3 x 3 px edge 
enhancement filter is used with the following weight 
factors 
1 1 1 
1-8 1 
1 1 1 
On the resulting values a threshold is applied to 
separate the lines+noise from the paper; in this 
example: 
black (lines+noise) 1 40 
while (paper) < 40 
Step 3 separation of boundary lines and noise 
Remaining noise and written landuse codes are now 
separated from the boundary lines using an AED 
polygon fill command. The cursor is moved on to 
point on a line and a polygon fill command to color 
the black line red is given. Eventually all lines 
connecting to that point will be colored red while 
disconnected linegroups, codes and noise remain 
black. In this stage gaps can be edited and the 
result is a bitmap of the boundary lines only. 
Ideally all pixels of the boundary lines should form 
a connecting network (except for the island polygon 
boundaries). To apply the AED polygon fill it is 
however important that this connectivity is in the 
form of "edge-connectivity", i.e. connected pixels 
should have at least one edge in common. 
To ensure this condition, a special 2x2 "edge 
connectivity" operator is applied prior to the 
actual separation. This operator recognizes cases of 
pixels that are only connected by one corner and 
adds an additional pixel to ensure that the pixels 
will be edge connected. The principle and some 
results of this operator are shown in fig. 6. 
Step 4 digitizing fiducial marks 
Fiducial marks are digitized with the AED cursor and 
their locations are stored to serve as reference 
point for later geocorrection procedures. 
Step 5 entering land use codes 
The operator can now enter the landuse codes by 
pointing with the cursor to an area, reading on the 
screen the (black) handwritten landuse code that was 
scanned together with the lines, and entering that 
code via the keyboard or a screen menu. The AED 
boundary fill is then used to give all pixels in the 
polygon the relevant landuse code. The operator can 
follow this as the polygon will be filled with the 
appropriate landuse code on the screen. In this 
interaction color is essential to minimize coding 
errors. 
Results of the polygon encoding are shown in fig. 7. 
An alternative is to digitize landuse centroids on a 
normal manual digitizer and use these points to fill 
the polygons. This could be an off line or an 
interactive procedure and further tests will have to 
show which method will be the most efficient. 
Step 6 editing 
The same applies to editing procedures. Gaps and 
other mistakes may be corrected in step 3 or in step 
5 when a "leaking" polygon is detected. Depending on 
error types and user experience it may be preferred 
to correct the overlay and rescan it, rather than go 
through tedious editing procedures. On the other 
hand minor errors may be edited quicker and easier 
at the screen, while also a certain amount of 
automatic error correction may be carried out 
especially in cases where map contents and topology 
are well defined.