The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part Bl. Beijing 2008 
438 
Table 5 shows the coherences of different oversampling 
methods: oversampling cross-correlation function vs. 
oversampling SAR images. 
Searching Subpixel Tie Points and 
Offsets 
# of 
Pars 
Interpola 
tor 
Coherence 
Method 
Rate 
Kernel 
North 
South 
Fit Peak 
10 
Spline 
6 
Sinc4x4 
0.4514 
0.4217 
Up SAR 
10 
Sinc40 
6 
Sinc4x4 
0.4509 
0.4217 
Table 5 Coherences of different oversampling methods 
Oversampling the SAR image patches did not produce higher 
coherence than oversampling the cross-correlation function. 
Oversampling SAR images is also slower than oversampling 
the cross-correlation function. 
The coherence for these two pairs is between 0.4 and 0.5, which 
seems a little low. The other pairs could have higher coherence, 
depending on the SAR data properties. 
4.3 InSAR DEM Evaluation 
Due to the inaccurate orbit and baseline information, InSAR 
processing without accurate ground control points could result 
large error in the DEM. By looking at northern Indiana in Table 
6, one can find the InSAR DEM of bilinear has slightly smaller 
DEM error than sine interpolator, although sine interpolator has 
the higher coherence. It is reverse for southern Indiana data: 
sine interpolation has the lower coherence, while its InSAR 
DEM has much higher accuracy. 
Searching Subpixel Tie Points 
and Offsets 
#of 
Pars 
Interpolat 
or 
InSAR DEM RMSE 
(Meters) 
Method 
Rate 
Kernel 
North 
South 
Fit Peak 
10 
Spline 
6 
Bilinear 
220.64 
129.16 
Fit Peak 
10 
Spline 
6 
Sinc4x4 
221.80 
65.40 
Table 6 InSAR DEM RMSE 
5. CONCLUSION 
Computing cross-correlation with magnitude only is adequate 
for both coarse and fine coregistration of ERS SAR data with 
medium baseline, regardless of whatever terrain is: flat or hilly 
area. Oversampling the cross-correlation function is more 
efficient than oversampling SAR images for fine coregistration. 
The experiments verified oversampling by a factor of 10, and 
concluded that a particular 4-parameter transformation was 
sufficient for subpixel coregistration of ERS SAR tandem data. 
The widely used resampling algorithms: nearest neighbor, 
bilinear, and cubic convolution, were tested with those ERS 
SAR tandem data and were compared to the computationally 
intensive sine interpolators with varied lengths. The longer sine 
interpolator produced a fluctuating but rising coherence. The 
2D sine interpolation with windowing and modulation exhibited 
the power of preserving the frequency spectrum, though no 
evidence showed the sine interpolator to have better coherence 
than bilinear or cubic convolution. 
This study indicates there may not be a best interpolator for 
resampling SAR images for all situations. The resampling 
preference can be affected by terrain type, SAR data type and 
quality. Coherence is not always a good criterion for estimating 
the resampling. It is a good indicator for evaluating 
coregistration within a single image. A higher coherence area 
indicates a better coregistration location, but coherence may not 
be a good indicator for comparing coregistration performance 
of different data sets, or different interpolators. The accuracy of 
the final InSAR DEM should be the ultimate standard for 
evaluating coregistration and the whole InSAR processing. 
REFERENCES 
ESA, E., 1999. Image Interpolation. Nuava Telespazio(3): 93. 
Gabriel, A.K. and Goldstein, R.M., 1988. Crossed orbit 
interferometry: Theory and experimental results from SIR-B. 
International Journal of Remote Sensing, 9(5): 857-872. 
Hanssen, R. and Bamler, R., 1999. Evaluation of interpolation 
kernels for SAR interferometry. Geoscience and Remote 
Sensing, IEEE Transactions on, 37(1): 318-321. 
Kwoh, L.K., Chang, E.C., Heng, W.C.A. and Hock, L., 1994. 
DTM generation from 35-day repeat pass ERS-1 interferometry, 
Geoscience and Remote Sensing Symposium, 1994. IGARSS 
'94. Surface and Atmospheric Remote Sensing: Technologies, 
Data Analysis and Interpretation., International, pp. 2288-2290 
vol.4. 
Leical, 2007. Introduction to Radar Data. Leica ERDAS 
IMAGINE Online Documentation, 9.1. 
Li, F.K. and Goldstein, R.M., 1990. Studies of multibaseline 
spacebome interferometric synthetic aperture radars. 
Geoscience and Remote Sensing, IEEE Transactions on, 28(1): 
88-97. 
Liao, M., Lin, H. and Zhang, Z., 2004. Automatic Registration 
of InSAR Data Based on Least-Square Matching and Multi- 
Step Strategy. Photogrammetric Engineering & Remote 
Sensing, 70(10). 
Lin, Q., Vesecky, J.F. and Zebker, H.A., 1992. New approaches 
in interferometric SAR data processing. Geoscience and 
Remote Sensing, IEEE Transactions on, 30(3): 560-567. 
Prati, C. and Rocca, F., 1990. Limits to the resolution of 
elevation maps from stereo SAR images. International Journal 
of Remote Sensing, 11(12): 2215-2235. 
Rufino, G., Moccia, A. and Esposito, S., 1996. DEM generation 
by means of ERS tandem data Proceedings of the Fringe '96 
Workshop ERS SAR, Zurich, Switzerland. 
Rufino, G., Moccia, A. and Esposito, S., 1998. DEM generation 
by means of ERS tandem data. Geoscience and Remote Sensing, 
IEEE Transactions on, 36(6): 1905-1912. 
ACKNOWLEDGMENTS 
ERS data were offered by the European Space Agency. Indiana 
high resolution DEM data were provided by [2005 Indiana 
Orthophotography (IndianaMap Framework Data 
www. indianamap. org)].