EM
ality
ved
nal
ted
ned
the
ing
the
trip
EM
wee
rain
ting
and
19s,
res.
Om.
0m
10S,
and
03.
that
have been investigated and the results which have been
obtained.
All the DEMs used in Tables 1 to 3 have been produced
on a 5m grid interval using the red colour band.
Automatically generated DEMs can be produced on
either the red, green or blue bands. Normally the red
band is considered to produce the best results.
Breaklines have also been digitised along the top and
bottom of the cliff for the six DEMs produced on the
ImageStation (is01-is06). At present, it is not possible to
digitise breaklines using the ERDAS software.
Both systems contain a comprehensive list of DEM
parameter settings which can be varied. The ERDAS
software contains 16 settings, while the ImageStation
has 28. Initial research into the production of automatic
DEM generation has concentrated on what are thought to
be the more influential parameter settings.
4.1 The DEM Parameter Settings
The ImageStation MATCH-T software generates the DEM
points essentially from feature mapping. |t identifies a
large number of individual image feature points that are
used for the subsequent derivation of the terrain surface
points. The ERDAS automatic DEM algorithm is
classified as an area correlator in which patches of pixels
from one image are correlated during the matching
process with conjugate patches on the other. The
primary correlation measure takes into account overall
differences in tonal contrast between the image patches.
The three ImageStation terrain setting options, flat, hilly
or mountainous can be influenced by a smoothing filter
which can be set at either high, medium or low. The
ERDAS DEMs have been varied using just three
parameters, namely the minimum and maximum
template sizes and the maximum parallax value.
Thus, the template size refers to the dimensions, in
pixels, of the correlation template. The software begins
each correlation for an area using the minimum template
size. If a successful correlation is not reached, then the
next larger size template is used and another attempt is
made. This is repeated until a match is made or until the
maximum template size is reached, in which case the
correlation attempts for the area are abandoned. The
template sizes are normally considered as dependant on
the image quality and content. In this coastal zone, the
cliff top is mainly low, undulating relief of green furrowed
fields. The beach area is essentially composed of fine
sand and small pebbles. In both cases, one area looks
almost identical to another i.e. low relief and low image
content.
The maximum parallax parameter is a function of the
elevation range in the model area. Each time a DEM
correlation is performed, the algorithm allows the search
area to be expanded above and below the predicted
elevation of a point by this amount.
Larger values of maximum parallax will search over
greater elevation ranges and smaller values often miss
points due to not finding the elevation within the more
restricted range. Separating the cliff top and the beach,
921
at Easington, is a cliff line approximately 17m to 22m in
height, which is high in image content, showing good
variations in tone and texture. Thus, an optimum
parallax value has to be obtained which allows for the
best modelling of the steep cliff face and the flatter areas
on the beach and cliff top.
4.2 Analysis of Results
The results in Tables 1 to 3 consider 9084 grid points
covering the beach, cliff, and cliff top. The ImageStation
settings used are shown in Table 1. A comparison
between flat, hilly and mountainous settings using a low
smoothing filter shows that the hilly setting is producing
slightly better results than the flat setting which is
producing better results than the mountainous setting.
This is particularly shown by the percentage of points for
the range +0.2m and the higher percentage of points
within the lower range, +0.05m for the hilly setting. The
hilly setting can also be compared with the flat setting for
a medium filter and again the hilly setting would be
considered better.
Looking at the effects of the smoothing filter, considering
the flat terrain setting where there are high, medium and
low filter results, shows the high filter producing results
perhaps just slightly better than the medium filter and
both are better than the low filter. Similarly for the hilly
settings the medium filter is producing the best results.
So the best settings for this model are the hilly, medium
filter settings. Possibly the best results might have been
achieved with a hilly, high filter setting.
If these results are compared with the theoretical values
calculated in section 3, the theoretical flat and hilly
values are probably somewhat optimistic. The
theoretical value for the mountainous setting would
perhaps be more appropriate for these two settings.
Terrain Flat Hilly Mts
setting
Smoothing | High | Med | Low | Med | Low | Low
filter
Range (m) | is01 | is02 | is03 | is04 | is05 | is06
Under -0.20 7851 681 81 5.6 48.1 9.5
-0.20>-0.15 281 32| 40! 32} 40) 329
-0.15>-0.10 48].5.4] 591|l 56l 60] 55
-0.10>-0.05 88) 8241.78 821.76] 7.3
-0.05> 0.00 10.6 | 10.81...8.6 |. 10.81. .8.7 |. 8.7
0.00<+0.05 14.1 1.10.7.1...9.9.].14.5 1.40.3.1.. 9.9
0.05<+0.10 | 10.9 | 11.1 | 10.0 | 10.6 | 10.4 | 9.2
0.10<+0.15 10.0; 10.0] 9.3| 10.4] 9.4] 8.9
0.15<+0.20 9.1 931190 92] 86. 80
Over 40.20 | 24.1 | 24.5 | 26.3| 24.9 | 26.8 | 29.0
10.05 21.7.| 21.5.] 18.5 ] 22.3 | 19.0 |. 18.6
10.10 41.4 | 40.8 | 36.3 | 41.4 | 37.0 | 35.1
+0.15 56.2 | 56.2 | 49.6 | 57.4 | 52.4 | 49.5
+0.20 68.1 | 68.7 | 64.6 | 69.5 ] 65.1] 61.5
Under -0.20 78 681 9.1 56] 81 9.5
Over 40.20 | 24.1 | 24.5 | 26.3 | 24.9 | 26.8 | 29.0
Table 1 ImageStation DEM Comparison Statistics
(96 of points)
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B4. Vienna 1996