Table 2. Mean, standard deviation and speckle index
processed by 8 filters.
Filter M.V. S.D. Speckle
Index
Original 187.54 57.94 0.309
Box 186.42 18.94 0.102
Median 186.84 19.08 0.102
Lee 186.42 18.94 0.102
Frost 186.44 19.17 0.103
Kuan 186.42 18.94 0.102
Enh. Lee 194.29 19.86 0.102
Enh. Frost 194.30 19.36 0.100
G-MAP 192.21 | 20.89 0.109
In order to test the performance of filtering techniques
for a different types of area, Table 3 is made below.
Several test areas were selected, such as ocean, coastal
line, etc. and testing results present the ability of
filtering homogeneous areas and heterogeneous areas.
The speckle index is chosen as a criteria to measure the
reduction of speckle noise by these filters. The
homogeneous and heterogeneous areas were measured
by the speckle index. From Table 3, for the
heterogeneous areas, the speckle index is relatively
higher. The GMAP filter is much more efficient for
reducing the noise while keeping the edges.
Table 3 Speckle index test for different terrain.
Filter From homo- to Hetero-
geneous geneous
Original 0.309 0.312 0.328 0.464
Box 0.105 0.121 0.142 0.228
Median 0.112 0.123 0.152 0.220
Lee 0.105 0.121 0.142 0.228
Frost 0.107 0.122 0.145 0.235
Kuan 0.104 0.121 0.142 0.228
Enh. Lee 0:112 0:125 0.152 0.328
Enh.Frost | 0.107 0.125 0.147 0.301
GMAP 0.126 0.128 0.167 0.384
Qualitative Evaluation:
The qualitative evaluation was made by visual
interpretation. A 3-look ERS-1 image with the size of
256 by 256 pixels was selected to test the performance
of the filters. The test results is shown in Fig.4. Fig4(a)
is the original image, which is an agricultural land area.
Fig4.(b) illustrates the Box Filter filtered image. The
speckle is reduced dramatically. However, the image
also became blurred. The Median filter is slightly
superior to the Box Filter for keeping edges(ref.
Fig.4(c)). The Lee filter improves the ability of
preserving edges compared with the box filter while
reducing speckle noise(Fig.4.(d)). The Kuan Filter and
Frost Filter are similar to the Lee Filter. Enhanced Lee
Filter and Enhanced Frost Filter make improvements on
reducing speckle noise at the edges. However, it cannot
remedy sharp spot noise. GMAP Filter ptoved to be
168
much better for filtering speckle noise while preserving
the edges. From a visual interpretation point of view, it
is the best one.
Fig.4, (a) Original ERS-1 3-look image.(b). Box Filter. (c).
Median Filter. (d). Lee Filter. (e). Frost Filter. (f). Kuan
Filter. (g). Enhanced Lee Filter. (h). Enhanced Frost
Filter.(i). GMAP Filter. ‘
In order to evaluate the performance of the filters with
different window size and different damping factors,
which are parameters available in the PCI software
package, we take the Frost Filter as an example. The
tested results are shown in Fig.5. From Fig.5, we
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B2. Vienna 1996
c
si
ec
ec
in
tec
res
de
an
Fil
fili
int
de