Geography Reference
In-Depth Information
Table 6.3
Advantages and disadvantages of commonly used image processing techniques to obtain flood area from SAR images
(After Di Baldassarre et al., 2011).
Image
Strength
Limitation
Level of
Computational
Level of
Consist-
Processing
Complexity
Efficiency
Automation
ency*
Method
Visual interpre-
tation
Easy to perform in case
of a skilled and
experienced
operator with
knowledge of flood
processes
Very subjective;
Difficult to
implement over
many images;
May be difficult for
images that show
complex flood
paths;
Low to high
(may have
varying
degrees of
complex-
ity)
Relatively low
Hardly
possible
>
0.9
Histogram
thresholding
Easy and quick to
apply;
Objective method;
No flexibility;
Optimized threshold
might not be the
most appropriate;
Works only well if
image is relatively
little distorted
Very low
Very high
Full
0.8
Texture based
Takes account of the
SAR textural
variation;
Based on statistics;
Mimics human
interpretation as it
takes account of
tonal differences;
Difficult to choose
correct window
size and
appropriate
texture measure;
After application still
requires threshold
value to obtain
flood area
classification
Moderate
Moderate
Full
0.6**
Active contour
modelling/
Region
growing
Image statistics based;
Usually provides good
classification results;
Easy to define seed
region (e.g. on the
river channel);
If integrated with land
elevation constraints
(see Mason et al.,
2007) results are
improved by
mimicking
inundation processes
Requires several
parameters to
fine-tune;
Slow on large image
domains;
Difficult to choose
correct tolerance
criterion;
May miss separated
patches of dry or
flooded land
(particularly in
case of active
contour
modelling)
Moderate to
high
Moderate
(depending
very much on
domain size)
Relatively
high
0.7
∗
Refers to consistency of binary classification between different SAR images, after Schumann et al. (2009a);
∗∗
Average of different texture measures
Search WWH ::
Custom Search