The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part Bl. Beijing 2008 
120 
Method 
Window size 3x3 
Window size 5x5 
Window size 7x7 
IV 
MSE 
SNR 
PSNR 
IV 
MSE 
SNR 
PSNR 
IV 
MSE 
SNR 
PSNR 
Original 
image 
540.04 
540.04 
540.04 
Lee 
540.04 
105.60 
17.076 
27.076 
540.04 
170.81 
14.932 
25.806 
540.04 
198.36 
14.249 
25.156 
Gamma 
540.04 
110.20 
16.874 
27.709 
540.04 
181.74 
14.656 
25.536 
540.04 
212.92 
13.945 
24.849 
Median 
540.04 
120.34 
16.384 
27.327 
540.04 
201.56 
14.006 
25.087 
540.04 
237.36 
13.228 
24.377 
Frost 
540.04 
104.14 
17.175 
27.955 
540.04 
162.65 
15.188 
26.018 
540.04 
190.25 
14.487 
25.337 
Table 1. Evaluation indices for standard filter 
4.5 Peak Signal-to-Noise Ratio (PSNR) 
N*255 2 
PSNR = 10.1o glo (_=r——-5-) 
2~t 2L v. 
i j 
PSNR is used for quantitative comparison. 
4.6 Signal-to-Noise Ratio (SNR) 
SNR = 10- log 10 (i 
I j 
-) 
>v' : 
i j 
4.7 Edge Preservation Index (EPI) 
^Ps&D-Psi*+U)| +\p s (i,j)-p s (i,j+X){) 
^Po&A-PoQ + U)| +\Po&j)-Po(U + !)|) 
(11) 
(12) 
(13) 
Here p s (i,j) is the denoising image, p 0 (i,j) is the original 
image. p s (i,j) and p s (i,j) are the pixel intensity of 
detected edges. The range of EPI is [0, 1]. 
5. EXPERIMENT AND RESULT ANALYSIS 
To verify the method proposed in this paper, experiments have 
been performed. The original image used is the airborne SAR 
image of one place in China, whose size is 512x512 pixels. 
In order to compare the optimal filtering results of each 
traditional filter, three different window sizes are used, and the 
obtained value of IV, MSE, SNR and PSNR is shown in the 
Table l.From the results of Table 1, we can easily find that: 
firstly, the image variance of the denoising image does not vary 
in all the filters. The increase in window size deteriorates the 
SNR and PSNR, which means the worse image quality. 
Secondly, the optimal filtering results of all traditional filters 
are obtained when the window size is 3x3. So later, the results 
of filtering algorithm proposed in this paper is compared with 
them. 
(e) (f) 
Fig.2 (a) Original imagery of airborne SAR with 512x512 
pixels (b)-(f) filtered images by Lee, Gamma, Median, Frost 
and our filter respectively. 
To evaluate the speckle reduction ability as well as the ability 
of edge preservation, index like mean value, standard variance, 
RSV, ENL, PSNR and EPI are calculated for comparison. The 
result of our filter and other traditional filters is shown in the 
Table 2. The filtered images by the Lee, Gamma, Median, Frost 
and our filter are shown in Fig.2 (b)-(f). 
The edge detection result of original image and the image 
filtered by our algorithm is shown in the Fig.3. The