In: Wagner W., Székely, B. (eds.): ISPRS TC VII Symposium - 100 Years ISPRS, Vienna, Austria, July 5-7, 2010, IAPRS, Voi. XXXVIII, Part 7B 
645 
In SAR data, additionally the interferometric phase information 
is sensitive to surface changes and ground motion. The 
coherence, which expresses the complex correlation of two 
SAR scenes, can be used as an indicator of changes (Scheuchl 
et al., 2009; Wright et al., 2005). However, since the phase is 
sensitive to sub-wavelength changes, the information is related 
to very subtle changes, especially for X-Band. Therefore, 
change detection based on the interferometric coherence is 
susceptible to high false alarm rates, since several impacts, like 
e.g. weather and surface properties, can cause decorrelation. 
Incoherent and coherent change detections provide 
complementary characterisations, since they are sensitive to the 
different measures of a SAR scene: backscatter intensity and 
phase. Therefore, the joined use of both statistics often provides 
a better description of surface changes. Preiss et al., 2006 
showed a statistical test for change detection combining 
incoherent and coherent change statistics. 
To respond to certain customer-specific applications, the 
detected changes have to be assessed in a fast and efficient way. 
However, the automatic assessment of changes is a challenging 
task. Classification of the scene gives information about surface 
characteristics (Wegmiiller et al., 2003), which can be used to 
restrict the search areas and reduce unrequested changes. 
Additionally, this information gives an indication about the type 
of change. 
3. CHANGE ANALYSIS - DETECTION AND 
ASSESSMENT 
To detect changes, the different parameters of a complex SAR 
signal can be exploited, see Sec. 2. In the suggested process 
further information is integrated to decide whether the change is 
of interest or not for the customer-specific case. 
The following sketch shows a brief processing overview for the 
automatic detection and assessment of changes. 
Figure 1; Processing scheme - change detection and 
assessment 
3.1 Change Detection 
In our work we use both, the intensity ratio and the coherence 
as change indicators. We assume certain distributions of those 
features to estimate the thresholds to distinguish changes. The 
known distribution functions of the intensity ratio (Touzi et al., 
1988) and of the coherence (Hanssen, 2001) are used to 
separate changed from non-changed pixels, assuming a certain 
false alarm rate. 
The intensity ratio contains, not only the information that a 
change happened, but also if the backscatter increased or 
reduced between the acquisition dates. This information will be 
kept and can also be used in the following assessment step. 
3.2 Assessment of changes using additional information 
Into the proposed scheme further information for the assessment 
is integrated resulting in a context based change detection 
approach. Broad land cover classification of the pre-event scene 
is used as context information. 
Since there is not a unique descriptor applicable to various 
surfaces, different textures and statistics are estimated to 
improve land cover discrimination. Texture measures provide 
information on spatial variation of the backscatter and thus 
information of the local surface characteristics within the scene. 
This information together with coherence and backscatter 
statistics is used in a classification approach to restrict the 
search for areas and changes of interest and for the evaluation 
of changes. 
In a first attempt a straight forward pixel-wise classification 
based on coherence and backscatter intensity has been tested 
and achieved in general reasonable results concerning broad 
land cover classes such as urban, forest, open land, water and 
coastal areas. However, combining the classification result with 
the change indicator showed that the pixel-wise classification 
approach is too noisy to reduce unrequested changes. 
Therefore, object based classification methods have to be 
applied in order to first divide the image into homogeneous 
regions (segmentation) and second, derive image characteristics 
such as co-occurrence texture measures, coefficient of 
variation, and class related features from resulting image 
objects. Based on these features a simple decision tree can be 
set up to classify the pre-event scene. 
This kind of information gives a hint about the type of change. 
Additionally the change direction of the backscatter can be used 
to evaluate if an object disappeared or showed up. Hence it 
supports the classification of the type of change. 
4. APPLICATION TO DATA 
The combined detection and assessment of changes has been 
applied to TerraSAR-X StripMap acquisitions to demonstrate 
the approach. 
4.1 Available Data and Scenery 
Two TerraSAR-X StripMap (3m resolution) acquisitions are 
available. The images are repeat-pass acquisitions with the 
following acquisition parameters: HH polarisation, ascending 
orbit, incidence angle: 36-38,6 °. 
The scene is located in Panama close to the capital and the 
southern entrance to the Panama Canal. The following Figure 2 
and Figure 3 show only a detail of the StripMap scenes with an 
airport in the south western part, the entrance to the Panama