The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part В 7. Beijing 2008 
Where, |^| is the coherence level; Sj(/, y) , S 2 (i,j) are 
echo signal of master and slave image; * is the conjugate of a 
complex; M X TV is the window size during the coherence 
estimation. 
The interferogram phase can be computed by 
y/ = arctg(s x Vs*) (2) 
4. CASE STUDY 
4.1 Data sets 
Shanghai city, located in the east of china is selected as our 
experimental district. The SAR images and its location which 
are used to generate the interferogram covering Shanghai is 
showed in the Fig. 1, Tab. 1, and Tab. 2. 
Figure 1. The SAR image of Shanghai test site imaged on Jun, 6 
1992 
Mission Northwest Northeast Southwest Southeast 
Longitude 31° 19' 31°17' ЗГ06' 31°09' 
Latitude 121 °23' 121 °36' 121 °33' 121 °21' 
Table. 1 The Geographic coordinates of SAR test area 
Mission 
Master 
Slave 
Д T 
(d) 
B1 
(m) 
Spectral 
overlap 
ERS-1 
19920606 
19920919 
105 
-109.47 
Range : 89.84% 
Azimuth: 
95.96% 
Table. 2 The basic parameters of the SAR interferometric pairs 
of Frame 2979 
4.2 Comparison of several adaptive filtering methods 
Filtering methods based on statistical model are the most of 
generally used methods in the image processing due to its 
obvious advantage of reduction of the standard derivation. And 
it proves that this series of filter not only filter the noise 
preferably but also preserve the edge for the better. 
In our test algorithms such as Enhanced Lee, Enhanced Frost, 
Gamma Map, and Kuan filtering are used to remove the noise of 
the master and slave SAR images. To evaluate the efficiency of 
those methods, we select five criterions including SD (Standard 
Deviation), SI (Smoothness Index), SNI (Speckle Noise Index), 
ENL (Equivalent Number of Looks), and NM (Normalized 
Mean), seen in Tab. 3. For a quasi-homogeneous scene, SNI is 
an index which reflects the intension of speckle. The large SNI 
value means the high speckle noise level. In addition, the larger 
the SI and ENL, the stronger smoothness ability of the filtering 
algorithms is. The NM is more close to 1, the unbiased 
estimator is more approached. 
By comprehensively considering the above assessment indexes, 
the Gamma Map filtering is verified the optimal filtering due to 
its highest ENL and SI, and smallest SD and SNI. The next step 
is to generate the coherence and interferogram using the 
unfiltered pair and filtered one, respectively. 
Master 
19920606 
Original 
Enhanced Lee 
Enhanced 
Frost 
Gamma 
Map 
Kuan 
SD 
1547.7262 
826.0680 
793.7859 
776.5155 
1044.2183 
SI 
1.3487 
2.5189 
2.6123 
2.6900 
1.9990 
SNI 
0.7145 
0.3970 
0.3828 
0.3718 
0.5003 
ENL 
1.8169 
6.3449 
6.8241 
7.2361 
3.9960 
NM 
1.0000 
0.9968 
0.9934 
1.0007 
1.0000 
Table. 3 Comparison of Four kinds of filtering algorithms 
4.3 Interferometric data processing 
The main steps in our interferometric test include coregistration, 
removal of the flat earth phase, generation of interferogram and 
coherence, filtering of the interferogram. 
To reduce errors during the process of co-registration and flat 
earth phase removal, we used the precise ERS orbit data from 
Delft University of Technology. Consequently, we need execute 
general coarse to fine coregistration between the master and 
slave images. The final fine coregistration errors of unfiltered 
and filtered pairs are both controlled in the range of 0.05 pixels 
along the range direction and 0.08 pixels along the azimuth 
direction. In order to compare the interferometric results under 
the same condition, we adopted the same window size during 
the estimation of coherence. To suppress noise, the SAR images 
are processed by a multi-look operation of 5 pixels along the 
azimuth direction to obtain a final resolution of 20><20m, and 
filter the interferogram with Gaussian filtering. 
The interferogram and coherence are showed in Fig. 2, Fig. 3, 
Fig. 4, Fig.5, respectively