image 
ies in 
at 20 
sue is 
are 
; have 
in the 
/ over 
acy of 
been 
id with 
portant 
rithms 
ysical 
k and 
ture of 
ed for 
pair of 
> used 
terrain 
1s with 
Jenous 
micron 
mono 
/, both 
anned. 
micron 
| at 20 
IRSC's 
digital 
of 576 
for the 
nly two 
od in a 
ture of 
magery 
over a 
nber of 
elect a 
system 
ists Or 
logy, to 
vailable 
ck-box’ 
before 
and 17 
ccept a 
itomatic 
generation software is known as MATCH-T, which itself 
has 28 variable parameters. 
With these numbers of parameters the possible variations 
are almost endless, so any structured testing of their 
effects must be focused, methodical and extensive. 
Besides the large number, it is difficult to identify from 
their title or from the manual exactly their meaning and 
effect, even with knowledge of image matching 
techniques. The two techniques follow slightly different 
methods, the explanations of which are beyond the scope 
of this paper. For suitable information on various 
correlation techniques the reader is referred to Lemmens 
(1988) and Krzystek (1991). 
5. DEM ANALYSIS 
The method adopted to analyse the quality of the DEMs 
involved the comparison of the automatic DEM with 
respect to the current production procedure i.e. semi- 
automatic generation on an analytical plotter. The 
coincidence between the automatic and semi-automatic 
generations is seen as the initial test of precision and the 
specific use of the DEM as the definition of the required 
accuracy. Differences in DEMs produced with changes in 
the parameter settings were investigated. From these, 
height anomalies were obtained at each node of the grid 
which are then statistically analysed. 
6. RESULTS 
Work is still being undertaken on DEM quality analysis, 
but the following results show some of the work to date. 
The results are split into two sections which reflect the 
instrument used. Section 6.1 is concerned with the 
ERDAS experimentation and Section 6.2 focuses on the 
ImageStation results. Each of these sections is further 
Split into sub-sections reflecting the topography of the 
imagery: rural and urban. The results highlight the 
variations obtainable by altering individual parameters. 
The results presented are only part of an extensive 
program of investigation and have been chosen to give 
an insight into the potential differences which can occur 
rather than to achieve the optimal quality of result. 
6.1 ERDAS 
6.1.1 Rural 
The nature of the imagery was such that the DEMs 
Produced were slightly larger in size than what was 
required. This was due to the stereopair not being 
oriented to north. Effectively the software creates a best 
fit northerly oriented boundary around the actual defined 
grid boundary and generates points to this Secondary 
boundary. The DEMs thus generated and used in the 
comparisons contain points which were unrequested. 
These points may have an effect on the statistics 
presented, but continuing research will ascertain this 
effect. This point clearly highlights the fact that what is 
l'equested is not always provided. 
359 
The first option the user is prompted for when starting a 
DEM is the choice for the Source of the imagery. It may 
appear to be quite unimportant, but results suggest 
otherwise. There are two options; Block or Stereopair. 
Even with only one pair of images, both options are 
available and they produce considerably different results, 
as shown in Figure 1. There are a large number of points, 
8.8%, with + 2m difference. It is clear that this seemingly 
insignificant ^ parameter does make considerable 
difference to the resultant DEM. The manual explanations 
of the effect of these options is unclear. 
With colour imagery there is the choice of which colour 
band to use to perform the correlation. Generations with 
each band were compared with each other. Figures 2 and 
3 show some of the statistics. 
2500. 
   
   
  
  
  
2000} 
£15004 v 
© um Eum 
s C : 
© . : D 
e 100044. — 
500 | — B... ii 
o] a (de ps : HR. B B 
S Ph A Lg 2 es wm om 
EE à 555 558 
BE À 
m om n m S unio mn 3 m» 
= Un c Un c 
Height Difference in Metres 
Fig.1 Histogram of Height Differences 
Stereopair - Block (Rural) 
1800 — 
ico p 
1400) 
4. 1200} 
'e 1000 J Ed SE a 
e T. 
= 8. 
S 
Z 600. 
400 
  
2004 
  
02-2pun kamen | 
1 1 1 1 > e - - > 
DAN cC nu S uu» += 
=> UOS tA ° 
VIMUS eMe 
V V V V 
1 1 1 ec e — - v ® 
nime S. mesa in 9 2 
C Ses tA e 
Height Difference in Metres 
Fig. 2 Histogram of Height Differences 
Band 3 (Blue) - Band 1 (Red) 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B2. Vienna 1996