Figure 5 Coherence map with Complete 
Average filter 
  
Figure 6 Coherence map with Incomplete 
Average filter 
The following process is developed to improve the 
resolution of the coherence map. 
l. Compute the coherence map with an average 
filter where each sub-image element gives the 
same contribution. 
2. Compute the coherence map with an average 
filter where each sub image element gives the 
same contribution except the center element 
of that sub-image. In this case, the center of 
the kernel is zero. 
3. Compute the normalized master and slave 
images. 
4. Compute the coherence map of the 
normalized images. 
5. Compare the elements of the coherence maps 
that are created with the two average filters. 
The two coherence maps that result from the two 
filters are usually different, especially at those 
points where the backscatter differs significantly 
from the surrounding. In practice this will mainly 
appear in points with a high backscatter. The 
following steps are executed to finalize the process. 
6. The difference in coherence between the two 
maps is computed. : 
7. If that difference is larger than a given 
threshold, then the coherence value that is 
computed with the complete average filter is 
assigned to the final coherence map. 
Otherwise, the final coherence map gets the 
value that is computed using the normalized 
images. 
The absolute difference between elements of the 
two coherence maps will easier exceed the 
threshold in areas with an overall low coherence 
than in areas with a relatively high coherence. To 
avoid an imbalance in the assignment of the final 
coherence, it is required to adapt the threshold to 
the local situation. It is also possible to add another 
component in the computation of the value to be 
tested. In this research, the last method is applied. 
The test value is computed as the product of the 
coherence value, applying the complete average 
filter and the difference between the two coherence 
maps. The method used can be described briefly in 
the following way: 
* Ciis the coherence value that is computed 
with the complete average filter. 
* Ciis the coherence value that is computed 
with the average filter of which the kernel 
center is equal to zero. 
*  C,is the coherence value that is computed 
from the two normalized images. 
e Cis the final coherence map. 
e © is the threshold 
The following statement is used for the final 
assignment: 
If o <C,*C, THEN C;=C, ELSE C;=C, 
Intemational Archives of Photogrammetry and Remote Sensing. Vol. XXXII, Part 7, Budapest, 1998