29 
tical data. IEEE 
Symposium,613- 
ation of Different 
, 2003. 
metric for image 
rence on Image 
on measure for 
ïtters, 38(7):313- 
al image quality 
-84, 2002. 
ORTHORECTIFYING SPACEBORNE SAR BY DEM BASED ON FINE REGISTRATION 
Hongjian You 3 Fu Kun a 
a Institute of Electronics, Chinese Academy of Sciences, Beijing 100080-youhongjian@263.net 
KEY WORDS: DEM, simulated SAR, orthorectification, TIN 
ABSTRACT: 
A method to orthorectify spacebome SAR based on DEM is described in this paper. DEM data is used to generate a simulated SAR 
image according to range-Doppler principle and experiential backscatter coefficient, and then we apply Harris operator and mutual 
information matching to extract homologous points distributed in real SAR image and simulated SAR image. Real SAR image can 
be registered finely with simulated SAR image based on triangle irregular network (TIN) that is constructed using the extracted 
homologous points. The real SAR image is then orthorectified based on the relationship between real SAR and simulated SAR, as 
well as the relationship between simulated SAR and DEM. The method is used to process Radarsat-1 SAR image and the processed 
result is satisfactory. 
1. INTRODUCTION 
Being an effective earth observation technique, Synthetic 
aperture radar (SAR) is being paid importance to and it is 
complimentary to optical remote sensing owing to its all- 
weather capability. Several satellite SAR systems have been 
launched since the first space-bome SAR run in 1978 and 
Chinese first space-bome SAR used for environment 
monitoring will also be launched soon. The space-bome SAR 
image must be ortho-rectified in order to provide an image 
without any distortion before it is widely used. Generally 
digital elevation model (DEM) is used to eliminate the 
influence of terrain, especially in mountain areas. There are two 
methods to rectify SAR using DEM (Johnsen,1995), one is to 
fine the satellite orbit more accurately using ground control 
points (GCP) that is collected in topography map and then the 
DEM is combined with fine orbit parameters to rectify the SAR 
image. The other one is to generate a simulated SAR image 
using DEM directly, and then the real SAR image is rectified 
into the geo-referenced coordinates of DEM after the 
corresponding relationship between the simulate SAR image 
and real SAR image is determined by matching manually or 
automatically. The latter method is an effective method because 
GCP is not needed and only DEM is needed, therefore it is 
possible to ortho-rectify SAR image automatically. The local 
distortion cannot be considered using the former method, and 
the simulated SAR image must be registered finely with real 
SAR image in order to eliminate the local distortion if the latter 
method is used. A new method to register the simulated SAR 
with real SAR from registration register to fine registration is 
proposed in this paper, and then space-bome SAR is ortho- 
rectified based on finely registered small region. 2 
2. SIMULATING SAR IMAGE BASED ON DEM 
Simulated SAR image can be generated based on SAR imaging 
geometry principle using the satellite orbit parameters and 
DEM. Both the image coordinates and gray-value 
corresponding to the grid location of DEM must be calculated 
in order to generate a simulated SAR image. We can use the 
range-Doppler formula to solve the image coordinates of DEM 
grid point because the SAR image is slant image which is 
synthetic using Doppler principle. The gray-level of SAR must 
be also simulated in addition to image location. The backscatter 
coefficient influences the gray-level of SAR and it is 
determined by wavelength, polarization, local incidence angle, 
coverage of vegetation, roughness of terrain and so on. Now it 
is very difficult to obtain accurate backscatter coefficient, so 
some physical models are proposed through theory analysis 
(Zhang, 2001), and some experiential formulas are also 
proposed to fit the backscatter coefficient through surveyed 
data in field. Our goal to simulate a SAR image is to ortho 
rectify the SAR, so the texture and geometrical features are key 
factors in order to match the simulated SAR with real SAR 
image and the absolute gray-level value is not important, 
therefore we use quadratic polynomial to calculate the gray- 
level of SAR image based on the gradient of ground because 
the gray-level of the SAR image correlates local incidence 
angle while the local incidence angle correlates ground gradient. 
The simulated image must be interpolated by neighbour DEM 
grid points because the image of hillside toward the SAR side 
look direction shrinks while that along the side look direction is 
sparse because of undulation of terrain. Figure 1 shows the 
simulated SAR image based on DEM and we can see that it is 
very similar with the real SAR image in figure 2, so it is 
possible and feasible to simulate SAR using DEM. 
Figure 1. Simulated SAR image