EEE SL
@
oO
>
$ 10%
5 -
8 EH RN AT ;
cc 8r i
o
c
uj aa’ 1 FO OF lE |
e 8 re fen rn: enn dex
&
4. f
L4 CONT- 100 -a—
ok CONT= 40 --— -
CONT- Mean -*--
L 1 1 1 1
0 À 1
EPOS GEOM R-LEE E-FRO FRO KUAN LEE E-LEE G-MAP
Figure 3: Distortions at different contrast values (Areas)
since an edge with a high contrast may be detected much
better. For low contrast values the geometric filter seems to
perform good for areas and lines, but even bad for points and
according to the mean retention.
Finally the computation time needed on a sun sparc 20 for the
test image was measured. Most filters are within a close field
of 30 — 70 seconds. Only R-LEE (112 sec.) and the GEOM
filter (186 sec.) need more computation time since more than
one iteration has to be performed. Since computer hardware
performance increases very fast, we regard the computation
time as a secondary criterion for a rating of the filters. The
computation time of one ERS-1 scene at a 25 meter resolu-
tion is approximately 6 hours for the GEOM filter, what is
practicable for most applications.
6 SUMMARY
A method has been presented to compare the performance
of adaptive speckle filters in a more objective manner as it is
done by other review papers so far. Therefore the smoothing
capability of all filters in homogeneous areas was adjusted to
a similar high level, as it is necessary to achieve a practicable
radiometric image quality for several applications. On this
basis the geometric distortions at different geometric primi-
tives was measured by the RMS-error from a synthetic image.
The method was applied to images at different signal levels
with respect to the amount of speckle noise, thus we got
an exhausting overview of the performance of various filter
algorithms.
In general, algorithms which take into account geometric as-
pects, as the EPOS, GEOM and R-LEE filter, achieve the
best overall performance, leaded by the EPOS filter. This is
due to the evaluation of information, contained in two dimen-
sional image signals by the strong embedding of each pixel
in its environment, as mentioned in Section 3. Especially for
signals at high contrast levels the EPOS filter outranges the
other methods clearly, at low contrast levels some other fil-
ters perform somewhat better. Especially the GEOM filter
seems to be preferable for the evaluation of areas and lines at
low contrast levels, if the retention of the mean value is not
necessary.
REFERENCES
Crimmins, T. (1985). Geometric filter for reducing speckle.
In SPIE, International Conference on Speckle, volume
556, pages 213-222.
Frost, V., Stiles, J., Shanmugan, K., and Holtzman, J.
(1982). A model for radar images and its application
to adaptive digital filtering of multiplicative noise. IEEE
Transactions on Pattern Analysis and Machine Intelli-
gence, 4(2):157-166.
Hagg, W., Segl, K., and Sties, M. (1995). Classification of
urban areas in multi-date ERS-1 images using structural
features and a neural network. In Proceedings IEEE In-
ternational Geoscience and Remote Sensing Symposium.
Hagg, W. and Sties, M. (1994). Efficient speckle filtering of
SAR images. In Proceedings IEEE International Geo-
science and Remote Sensing Symposium, pages 2140-
2142.
Haralick, R. (1978). Statistical and structural approaches
to texture. In Proc. of the 4. Int. Conf. on Pattern
Recognition, page 45.
Kuan, D., Sawchuk, A., Strand, T., and Chavel, P. (1985).
Adaptive noise smoothing filter for images with signal-
dependent noise. /EEE Transactions on Pattern Analysis
and Machine Intelligence, 7(2):165-177.
J. (1981). Refined filtering of image noise using local
statistics. Computer Vision, Graphics, and Image Pro-
cessing, 15:380—389.
Lee, J.-S. (1980). Digital image enhancement and noise
filtering by use of local statistics. IEEE Transactions
on Pattern Analysis and Machine Intelligence, PAM1-
2(2):165-168.
Lee, J.-S., Jurkevich, l., Dewaele, P., Wambacq, P., and
Oosterlinck, A. (1994). Speckle filtering of synthetic
aperture radar images: A review. In Remote Sensing
Reviews, volume 8, pages 313-340, USA. Harwood Aca-
demic Publishers.
Lopes, A., Nezry, E., Touzi, R., and Laur, H. (19902). Maxi-
mum a posteriori filtering and first order texture models
in sar images. In Proceedings IEEE International Geo-
science and Remote Sensing Symposium, pages 2409-
2412.
Lopes, A., Touzi, R., and Nezry, E. (1990b). Adaptive speckle
filters and scene heterogeneity. IEEE Transactions on
Geoscience and Remote Sensing, 28(6):992-1000.
Shi, Z. and Fung, K. B. (1994). A comparison of digital
speckle filters. In Proceedings IEEE International Geo-
science and Remote Sensing Symposium.
Ulaby, F., Moore, R., and Fung, A. (1982a). Microwave
Remote Sensing: Active and Passive. Remote Sensing.
Artech House, Inc., 685 Canton Street, Norwood, MA
02062.
Ulaby, F., Moore, R., and Fung, A. (1982b). Microwave
Remote Sensing: Active and Passive, chapter 7, pages
464-483. Volume Il of (Ulaby et al., 1982a).
Lee
140
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B2. Vienna 1996
KEY
tem,
ABS
The
cam
ing.
conc
We
spec
mult
For :
and
for c
tive
of di
mati
on lè
tems
one
land
tioni
ter, (
sens
tems
rada
ous
or re
With
satili
are
for n
port:
vent
The
Here
tions
for s
spat
spec
vege
are |
The
ope: