in order 
value of 
between 
quality 
e. Using 
EM was 
t copy 
ferences 
;M have 
nt closer 
f 0.18m, 
| bias in 
rass and 
erienced 
fine the 
e on the 
ficult to 
d digital 
imptions 
PIXEL 
irs were 
om four 
ges were 
ron. The 
ins and a 
sion was 
  
  
  
  
  
  
  
  
  
  
  
Size/m Type Pixel/m Mbyte Points match% Elapse System Exec. 
25/0.63 grey 21/13 1 80.0 94,102 99.0 25:42 30.6 1369.0 
50/1.25 grey 11/13.8 20.0 94,046 96.3 8:51 8.4 497.4 
50/1.25 col 11/13.8 60.0 94,853 98.3 11:04 22.4 513.4 
100/2.5 grey $/12.5 5.0 114,880 98.3 5:13 4.0 300.8 
200/5.0 grey 5/25.0 13 28,191 99.0 1:11 1.4 67.1 
400/10.0 grey 5/50.0 0.3 13,323 98.4 0:18 0.7 16.7 
  
  
Table 2, VirtuoZo Processing Time for SG Personal Iris 4d30 work station vs Scan Resolution 
where Size 
= pixel size in microns/size on ground in metres 
Type = greyscale or colour 
Pixels 
Size of the matching window in pixels/size on ground 
Mbytes = Size of the image files in mega bytes, minimum 
Points = Total number of points matched 
Mtch% = Mathematical reliability of match from VirtuoZo (not true accuracy) 
Elapse = Elapsed time for matching process (min:sec) 
System = System time for matching process (seconds) 
Exec = Execution time for matching process (seconds) 
Digital elevation models and contour maps were then derived 
by VirtuoZo for each of the resulting stereo coverages. Thus 
nine stereo pairs with a photo scale of 1:5000 were processed 
using VirtuoZo and joined to form a single DEM. . Similarly, 
further DEM's from four 1:10000 pairs and two 1:25000 pairs 
were also processed. The project area was finally covered by a 
section of the 1:50000 stereo model. Absolute orientation 
parameters were imported and kept identical for all tests in 
order to eliminate that potential operator induced error source. 
Processing times were recorded for the matching process in 
VirtuoZo to estimate the overhead caused by image size. Table 
2lists exemplary DEM processing times of VirtuoZo for one 
model 1:25 000, digitised at different pixel sizes. 
We note that timing took place on a slow Personal Iris 4d30 
workstation running at 30 Mhz, while normal production 
machines run in the order of 150-200 Mhz. It is interesting to 
note that the time change is non - linear. This is because the 
number of points matched is not dependant purely on the 
resolution but also on the number of pixels in the matching 
window. For this exercise we attempted to keep the size of the 
match window on the ground constant, however for the coarser 
resolutions we were unable to maintain this intention as the 
automatic process normally needs a minimum of 5 pixels to 
maintain its integrity. Absolute orientation parameters were 
imported for all tests in order to eliminate that potential 
operator induced error source of absolute orientation. 
Table 3 gives the accuracy results of the DEM tests, showing 
the dependency of height accuracy on scan size and photo scale. 
The RMS value of spot height measurements is given in units 
of metre. Also listed are the height accuracy specifications of 
the Queensland Department of Primary Industries (QDPI). We 
note, that VirtuoZo spot height accuracy is in the order of 1 
pixel for the small pixel sizes of 12.5 and 25 microns and 0.3 to 
0.5 pixel for pixel sizes 50 microns and beyond. It degrades 
approximately linear with photo scale. The results appear to be 
less dependent on scan resolution than on photo scale. The 
RMS values are in the order of 1/7000 of flying height for 25 
micron resolution, degrading to 1/2500 of flying height for the 
50 micron scans. No significant differences were found for 
colour and black/white photography. Based on the results 
available, we can conservatively estimate VirtuoZo's DEM 
accuracy to be in the order of 1 pixel size. As a word of caution 
we again point out that absolute orientation parameters were 
imported for all tests in order to eliminate that potential 
operator induced error source. 
4. COMPRESSION TESTS 
The 1:1600 images were further used to study the effect of 
JPEG compression (pixel size 25 micron) Orientation 
parameters were again imported in order to eliminate the 
orientation errors. Lossy compression was used with quality 
factors of 75 and 60. For one single image, the file size through 
this compression reduced from 81.8 Mb to 13.5 Mb and 10.6 
Mb respectively. The resulting DEM's showed differences of 
RMS 0.002m and 0.003m resp. against the uncompressed 
processing. This corresponds to 1 to 2 microns of accuracy loss 
and can readily be tolerated. 
Further compression according to the standard JPEG format 
created artefacts such as blockiness in the image which was 
regarded as unacceptable from a cartographic point of view, for 
the task of producing ortho photo maps. This blockiness is a 
clearly visible square block structure with noticeable different 
average grey levels in adjacent blocks (Wallace, 1991). 
205 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B2. Vienna 1996