f illustrates the results for a 3-look ERS-1 SAR image.
The results coincide with the theoretical analysis.
) However, the speckle index changes with the intensity
of the image. The lighter the image is, the smaller the
speckle index is. From this result, we indicate that the a
linear fit is not accurate enough.
] Probability density function(pdf) of SAR images: In
Xx] section 2, the statistical models of the SAR images are
] provided to have a Gamma distribution. Here we
process a patch of homogeneous area in the ERS-1 and
] ERS-2 image to test the distribution of the SAR images.
" The tested results is shown in Fig.2,(a),(b). Fig.2(a)
7 shows the pdf of a ERS-1 single-look image and
7 Fig2.(b) shows the pdf of a ERS-2 single-look image.
] The results accord with the Gamma distribution. The
{A slight deviation is due to the heterogeneous effect.
Distribution of ERS—1 Amplitude Rage
0.060 ETE e QW E RN VA UT 3
0.050 F
['rrTTTTFTT
< 0.040
X
T[TTTTTTTTTTTTT
3 0.030
0.020
0.010
EFTYTTT[YTTTTTTTT[TITTTTIT
0.000 CL Lene reed LT LT
uo 0 1.0x10* 2.0x10* 3.0x10* 4.0x10* 5.0x10*
Amplitude
7 Distribution of ERS-2 Amplitude imo
—À1 a 0.050E 777770777077 TITTY TT TTT TTT TTT TT TTT TTT TTT T
2.5x10 E
3 0.040 E
4 E
1 E
3 E + S
1 0.030 E- j x: d
my = E 1
1 a E 7
] 0.020 :; : =
jd 0.010 5 =
] UU OB de M. br E t VA ==> i
1 0 1.0x10^ 2.0x10* 3.0x10* 4.0x10* 5.0x10*
3 Amplitude
Quad Fig.2 (a). pdf of a ERS-1 single-look image
700
(b).pdf of a ERS-2 single-look image
Pdf of speckle index: It demonstrates the variation of
speckle index for any one SAR image and illustrates
the characteristics of a heterogeneous image. Fig.3(a)
he shows the pdf of speckle index for an ERS-1 image.
le- Several curves are overlapped in this figure to show
ne that the speckle index is changing with the variation of
I the terrain. The pdf of speckle index can be
its approximated as the Gaussian distribution for the
D homogeneous area. However, for the heterogeneous
C
167
area, the pdf of speckle index is a complicated function
and needs to be investigated (Ref. Fig.3 (b)).
0.20 7 ? : T , 7 , T ? à ; a
UU TUTUP T DEBT]
0.6 0.8
CTI ART NET TT OT B
0.06 —
0.04 LL +
0.00
0.6 0.8
Fig.3 (a) the pdf of speckle index for different areas
(b) the pdf of speckle index for a heterogeneous area
4. Evaluation of Filtering Techniques
Both quantitative evaluation and qualitative evaluation
have been made to evaluate the filtering techniques for
ERS-1&2 imagery.
Quantitative Evaluation:
The performance of evaluating eight filters for 3-look
ERS-1 images by computing the mean value, standard
deviation and speckle index is shown in table 2. The
window size of filtering is 7 by 7. The pixel number for
statistical value is 400. The testing was done in a HP
workstation using the PCI software package. From the
average statistical data, the Box Filter, Frost Filter, Lee
Filter, and Kuan filter have a good performance to keep
the mean value of the tested area. Median Filter,
Enhanced Lee Filter, Enhanced Frost Filter and GMAP
Filter make a small amount of change in the mean
value. Box filter is the best one to reduce speckle noise
for SAR images.
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B2. Vienna 1996