—— 
rferometric 
> and slant- 
mulation is 
the SNR is 
ghtforward 
;'oherencep 
(15) 
> SNR can 
(16) 
nall, 1994), 
e estimated 
the blurring 
data for the 
be taken in 
hting of the 
originating 
nages to be 
Still occurs 
surrounding 
ht variance. 
'ero weight. 
'am through 
J 
>d by phase 
s massively 
rier domain 
(17) 
    
  
  
Correlation, 
matching. 
Marc Honikel 
  
4 FUSION OF A SPOT STEREO DEM WITH AN ERS-1 INTERFEROGRAM 
41 Test data 
The test dataset consists of a pair of ERS-1 SLC quarter scenes (Frame 819, Quarter 1, Orbits 829 and 872), acquired in 
a three day interval (9/12/ and 9/15/91), and a SPOT stereo-pair, taken in a four day interval (9/3/ and 9/7/86). Both data 
sets are part of a data set of Catalonia used for our work in the EU project ORFEAS (Patias, 1998). 
The test site comprises an area of approximately 150kmY' and shows an undulating terrain with a height difference of 
315m between the minimal and maximal height. A DEM in 30m grid, derived from a 1:5000 topographic map, served 
as reference for our computations. The root mean square (rms) error of the reference DEM was approximately 1m. 
The data has been processed with commercial software packages. The SPOT DEM is part of the DEM of the full scene, 
which has been generated by the Leica Helava DPW 770 digital photogrammetric workstation, which uses cross 
correlation for matching. The correlation coefficient of each point is not accessible directly, but is hidden behind a 
figure of merit (FOM). This FOM is in a range from 0 to 100 and is related to the correlation coefficient (Leica, 1997). 
The SAR interferogram and the InSAR DEM have been generated with the PCI V6.1 IFSAR package. IFSAR offers 
also the opportunity for phase simulation from a given DEM. Both the original interferometric measurements and the 
results of the phase restoration are unwrapped with the ghost-line algorithm. Although the average coherence of the 
examined scene was relatively high (p = 0.5), due to the short repeat pass interval, several regions of low coherence 
have been observed and treated with the filter. 
Due to lacking ground control points (gcp), 5 gcps have been derived from the reference DEM for geocoding and 
baseline estimation improvement. Baseline fitting has been performed by a technique, proposed by Werner (1992). The 
simulated phases have been registered in slant-range to the geometry of the interferogram with 12 manually collected 
tie-points. 
42 Phaserestoration 
The Wiener filtering method has been applied to several regions within the interferogram, where the phase 
measurement initially failed. As the proposed method is spatially variant performing local filtering, certain regions of 
interest had to be found. As stated above, phase residuals, occurring due to noise or layover effects, are leading to phase 
unwrapping problems and were therefore targeted by the filtering. Residual reduction and the reconstruction of the 
fringe border are taken as measures of quality of the procedure. Residuals generally occur in presence of noise, hence 
coherence indicates the regions of interest to be treated with the Wiener filter. The coherence has therefore been used 
for the segmentation of the image in more and less affected regions. The filtering has been performed within a window 
of size 2" x 2" with natural numbers N and M required for the Fourier domain processing. Typical values for N and M, 
have been between 4 and 6. 
The measurement problems occurring in the interferogram proved the error assumption of the degradation model of 
section 1. Two main error types have been encountered and consequently removed: 
Fringe border degradation. 
Occurring mostly due to aliasing effects, it endangers the whole height determination process, forcing manual treatment 
in form of fringe line editing. In part, the problems in these areas have been so severe that even manual editing of fringe 
line was not possible. Fig. 6 shows an example of degradations of this type, where residuals (positive residuals are 
given the color green, negative residuals are red) and the corrupted fringe border (yellow) are shown. 
By applying the Wiener filter to these regions, the affected fringe borders could be restored completely. All occurring 
residuals vanished through the filtering, enabling smooth phase unwrapping even in those areas, in which the 
unwrapping originally failed. The fringe course could be estimated for those areas, where manual editing was initially 
not possible and had been excluded from the height determination process. As an example, the result of the filter 
operation, applied to Fig.6, is given in Fig. 7. 
Local noise. 
Noise occurs randomly within the fringes, due to decorrelation of the interferometric phases. These problems are less 
Severe compared to the problems caused by the fringe border degradation, as they are locally restricted. Still, as they 
occur randomly, they require detection by their low coherence value followed by local treatment. Fig. 8 shows a typical 
example of random local noise. 
The phase filtering performed also well with this type of degeneration. The affected areas have been smoothed 
according to the filter behavior in presence of additive noise. Adjacent valid fringe borders have not been affected from 
the filter operation proving the ability of the filter to adapt to the local noise (Fig. 9). 
  
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part Bl. Amsterdam 2000. 153