433 
IMAGE COREGISTRATION IN SAR INTERFEROMETRY 
Zhengxiao Li a , James Bethel b 
a ERDAS, Inc. (Leica Geosystems Geospatial Imaging), 5051 Peachtree Comers Circle, Norcross, GA 30092, USA - 
tonybest@gmail. com 
b Purdue University, School of Civil Engineering, 550 Stadium Mall Drive, West Lafayette, IN 47907-2051, USA - 
bethel@purdue.edu 
KEY WORDS: SAR Interferometry, Co-registration, Interferometric SAR (InSAR), Matching, Correlation, Interpolation, Synthetic 
aperture radar (SAR), Geophysics 
ABSTRACT: 
A critical procedure in Synthetic Aperture Radar (SAR) Interferometric (InSAR) processing was studied: SAR image coregistration. 
Two pairs of ERS-1/2 SAR tandem data, representing diverse terrain types and different baselines, were used in this research. The 
commonly used traditional SAR image coregistration algorithms were addressed and tested; the computationally intensive 
algorithms were examined; the results from those algorithms were compared, through the experiments carried out on real data. The 
results showed that the magnitude component had better performance compared to complex data for computing cross-correlation 
function. For fine coregistration, oversampling the cross-correlation function was more efficient than oversampling original SAR 
images and a factor of 10 was appropriate as the oversampling rate. A particular 4-parameter transformation was sufficient for 
subpixel coregistration of ERS SAR tandem data. The traditional resampling algorithms, nearest neighbour, bilinear, and cubic 
convolution, were tested and compared to the computationally intensive sine interpolators with varied lengths. The most efficient 
sine length was not always the longer one. The 2D sine interpolation with windowing and modulation demonstrated the power of 
frequency preservation, but no evidence showed that the sine produced better coherence than the common algorithms. The final 
InSAR DEM accuracy should be the ultimate standard for evaluating the best coregistration approaches. 
1. INTRODUCTION 
SAR interferometry requires pixel-to-pixel match between 
common features in SAR image pairs. Thus coregistration, the 
alignments of SAR images from two antennas, is an essential 
step for the accurate determination of phase difference and for 
noise reduction. SAR images are acquired from about 850 km 
slant range distance with baseline of approx. 200 m, so there is 
no visible parallax or disparity between the images. The entire 
purpose of the coregistration is to align the samples for phase 
differencing. The imprecise repeat-pass geometry makes 
coregistration difficult, and the InSAR complex data could 
facilitate coregistration. 
The normal optical image matching traditionally needs only one 
or two pixel accuracy, which is coarse coregistration for SAR 
images. The correlation window is used to search for offsets 
between master and slave images. After this pixel level 
coregistration, an interferogram may be generated, but it is not 
adequate for interferometric processing. The phase 
coregistration accuracy must be higher, so a subpixel level 
coregistration must be performed. 
The subpixel-to-subpixel match, also called fine coregistration, 
is a must for high precision InSAR results. Either the whole 
complex image or phase function is up-sampled to 1/8, 1/10, 
1/20, or even 1/100 pixel, in order to find the best sub-pixel 
alignment. One offset is not adequate for resampling the 
coregistered slave image. First or second order polynomial 
transformation equations are preferred to fit the conjugate 
matching points. 
The most commonly used method for coregistration is to 
compute the complex cross correlation function between the 
two SAR images (Li and Goldstein, 1990). Another approach 
involves estimating a signal-to-noise ratio (SNR) of the 
interferogram image (Gabriel and Goldstein, 1988). The 
average fluctuation function of the interferogram image can be 
used to adjust the coregistration parameters (Lin et al., 1992). 
The capability of computing hardware has advanced 
significantly in the past decade, while the cost has decreased 
tremendously. Advanced algorithms can now be used and 
improved operationally. 
In this paper, the commonly used SAR image coregistration 
algorithms are summarized, from the basic approaches, such as 
bilinear interpolation, to advanced methods, such as sine 
interpolation. Those algorithms were not only discussed 
theoretically but also examined with real data. The conditions, 
factors, and characteristics for those algorithms were analyzed 
and compared broadly. Enhancements to these algorithms were 
proposed and tested with real data. The experiments also 
evaluated the theory and simulations in the earlier papers. 
2. METHODOLOGY AND ALGORITHM 
The typical SAR coregistration procedure consists of 1) coarse 
coregistration for pixel level accuracy, including searching for 
coarse image offsets and shifting the slave image; 2) fine 
coregistration for subpixel accuracy, including searching for 
subpixel tie points, fitting transformation equations, and 
resampling the slave image. Coarse coregistration is a process 
to match two SAR images at up to one or two pixel accuracy. 
Fine coregistration is a process to find subpixel tie points on 
two SAR images, to fit transformation equations onto these tie 
points, and to resample one of these two SAR images based on