The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B7. Beijing 2008 
Figure 4. The image that range direction is from south to north 
is geometric corrected and subset 
3. SPECKLE REDUCTION 
The presence of speckle hinders human observer from analyzing 
the SAR image. Hence various method of speckle reduction is 
put forward. Although conventional speckle filtering algorithms 
can reduce speckle noise effectively, but also smear edges and 
blur images to some degree. The most well-known and widely 
used image-domain speckle filter is the local statistics adaptive 
filter proposed by Lee, which uses local statistics such as mean 
and standard deviation on fixed size window to determine the 
degree of smoothing (Yongwei Sheng,et al,J.). Although the 
Lee filter can preserve steep edges, the loss of fine details and 
degradation of spatial resolution may occur by using too large a 
window. On the other side, the use of a small window implies 
insufficient speckle noise suppression in homogeneous area. To 
solve the trade-off between the window size and the degree of 
speckle noise suppression, the article utilizes the adaptive 
window algorithm to reduce filter. The adaptive windowing 
algorithm was proposed to overcome the limitation of 
conventional image- domain speckle filters that have fixed size 
window( Dae-Won Do, et al., J., 2002).The steps are as follows: 
Set the max and min of window size, set the threshold of 
coefficient of variance; 
Calculate the sample mean and the sample variance of the 
boundary samples of current window; 
The sample mean is given by 
*e = 
4 * Chi “ l) 
T y&o 
mu 
Where Ly is the window height or width (height=width), y(k, 1) 
means the magnitude of complex SAR data. 
The sample variance is given by 
J, 
■ Qfljttln 
Calculate the coefficient of variance at (i,j) is defined as 
C 
11 = 
Cl! 
nii| 
The size of next window is determined by comparing the 
C,.with the threshold value T, f as following equation 
fralnÎNi! -F 1. Na„] if C u 2 
if £ it '£ T v go to the second step, the boundary samples 
becomes green pixels; 
if C{( i 7-j apply the Lee’s local statistics filter to the 
current window. 
Figure 5. Adaptive window