The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Voi. XXXVII. Part Bl. Beijing 2008 
Figure3. Result of different filters 
(a) Median, (b) Lee, (c) Lee-Sigma, (d) Frost, (e) Gamma-MAP, 
(f) Wiener, (g) wavelet threshold, (h) ISPW 
In order to comprehensively evaluate and analysis the 
performance of filtering results in Figure3, aside from visual 
quality five statistical values are considered. They are 
radiometric resolution (RR), effective number of looks (ENL) 
(Wakabayashi H and Arai K, 1996), signal-to-noise ratio (SNR), 
peak signal-to-noise ratio (PSNR) and edge preserving ability 
(EPA). Table3 shows statistical results of filtering results in 
Figure3. 
In view of visual quality, all kinds of filters can suppress 
speckle noise to some extent, and instantaneously blur edge 
information of original SAR image to certain degree as well. 
Among these images, proposed method owns the best visual 
quality. It preserves more edge information while removes more 
RR 
ENL 
SNR 
PSNR 
EPA 
Original 
1.4689 
1.5920 
— 
— 
— 
Median 
1.2537 
2.2796 
14.2822 
22.7056 
0.4241 
Lee 
1.2384 
2.3786 
15.0376 
23.4610 
0.4035 
Lee-Sigma 
1.2840 
2.2393 
15.3985 
23.8219 
0.4567 
Frost 
1.3656 
1.9578 
21.0488 
29.4722 
0.7286 
Gamma-MAP 
1.2322 
2.4039 
14.7218 
23.1452 
0.3864 
Wavelet threshold 
2.5968 
0.5030 
2.3372 
10.7606 
0.2755 
Wiener 
1.2774 
2.2910 
15.2190 
23.6425 
0.4438 
ISPW 
1.2086 
2.0171 
15.7170 
24.1404 
0.6103 
Table3. Statistical result of filtering images 
speckle noise than others as well, especially in river section, 
which reflects more distinctly and clearly. In contrast, Frost 
filter gives the worst visual quality for it removes only a few 
number of speckle noise, making denoised image the most 
difficult to discern. 
Radiometric resolution (RR) is to measure scattering 
electromagnetic features of targets; and the smaller radiation 
resolution value is, the more effectively it removes speckle 
noise. Table3 illustrates ISPW provides smaller value of RR in 
comparison to the other methods, indicating a better ability to 
reflect object scattering electromagnetic characteristic. 
Effective number of looks (ENL) reveals the ability to suppress 
speckle noise, hence the bigger ENL is, the stronger ability it 
removes speckle noise. It can be seen from Table3 that 
compared to other filters ENL value of ISPW is not optimistic, 
which remains deserve further improvement. 
Signal-to-noise ratio (SNR) and peak signal-to-noise ratio 
(PSNR) are the same sort of criteria to assess denoising result. 
The bigger SNR and PSNR values are, the more useful 
information image contains, and the better the algorithm 
suppresses speckle noise. It can be observed from Table3 that 
proposed ISPW method gives bigger value of SNR and PSNR 
than those of other methods except Frost, however denoised 
images of Frost filter is obviously worse than that using ISPW 
method. The reason for this phenomenon is that Frost filter 
removes comparatively less noise, making a smaller 
denominator in SNR and PSNR expression and inadequately 
improving SNR and PSNR values. 
Edge preserving ability (EPA) represents ability to preserve 
edge information of original images. The closer EPA 
approximates 1, the better denoised images preserve edge 
information of original ones. Tables3 reveals that EPA value of 
ISPW is larger that that of others’ except Frost, however, EPA 
value of Frost filter inaccurately indicates its edge-preserving 
ability, because Frost filter leaves more noise unremoved from 
original image including information mistakenly detected as 
edge, thus EPA value is inappropriately improved. 
From above analysis, statistical values reflect some but not all 
characteristics of denoised results. Thus it is fairly necessary to 
firstly consider visual quality into comprehensive evaluation. 
3.3 Experiment II 
We choose four filters, which generally obtains much better 
results in experiment I in both visual quality and statistical 
values aspects, to operate on another SAR image, containing 
different sort of objects to further evaluate performance of 
proposed ISPW method. Original image and corresponding 
SPOT5 image are shown in Figure4, denoised images of 
different filters and their statistical results are shown in Figure5 
and Table4 respectively. 
Figure4. Original SAR image (a) and corresponding SPOT5 
image (b)