From the available data set an area of 2700x1487 pixels was 
selected as a test image and the number of the extracted vector 
objects was limited to the related test area - 610 objects (Figure 
1). 
  
Figure 1. Test data set. Post-event image overlaid with building 
layer from the vector map. Satellite image courtesy of Digital 
Globe (©Digital Globe 2010). 
In order to verify the final result of our change detection the 
ground-truth information containing the state of buildings was 
initially identified by a visual comparison of the test data. 
3. METHODOLOGY 
To start a comparative analysis of two different data sources it 
is important to elucidate the knowledge that can be gathered 
from the available data sets. The remotely sensed image is in a 
raster format storing data as a regular array of pixels whereas 
the vector data are composed of three basic elements: points, 
lines and polygons. Due to the distinctions the vector and raster 
data provide different types of information. The vector data 
model presents objects with well-defined boundaries and their 
coordinates whereas the raster data model describes a situation 
at this place. Thus, the information about a building contour 
geometry and location as well as the area inside the building 
contour can be extracted from the available data. In order to 
assess the integrity of the building contours we developed a 
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XXXIX-B7, 2012 
XXII ISPRS Congress, 25 August — 01 September 2012, Melbourne, Australia 
feature ‘Detected Part of Contour’(DPC) that characterizes, 
which part of the building contour can be detected in the 
remotely sensed image. For an investigation of the area within a 
building contour several features based on textural information 
can be analysed. 
The investigated images can have different quality depending 
on atmospheric and brightness conditions. In order to reduce the 
influence of such effects, the images have to be corrected or 
filtered. Consequently, before starting the feature calculation 
process a brightness normalization across the satellite image is 
performed using homomorphic filtering (Gonzalez & Woods, 
2002; Delac et al., 2006). This is a frequency domain filtering, 
which provides the suppression of low frequency variation due 
to the illumination by taking the logarithm of the image 
intensity (that is a product of illumination and reflectance) 
before a high-pass filtering. The homomorphic filtering 
facilitates both: image brightness normalization and contrast 
increase (e.g., building edges and fragments of destruction). 
Finally, a binary classification of building states and a 
visualisation of the result damage map conclude the change 
detection technique (Figure 2). 
  
  
  
  
  
  
Remotely 
: Vector map 
sensed image 
Filtering Object 
S selection 
  
  
  
  
  
  
  
[ Feature extraction | 
(DPC, Textural features) 
Y 
Classification 
Ÿ 
Result visualization 
  
  
  
  
  
  
  
Figure 2. Steps involved in object-based change detection. 
3.1  Detected Part of Contour 
The ‘Detected Part of Contour’ (DPC) feature was developed to 
evaluate the integrity of building contours in the post-event 
images. If the investigated building has an intact contour the 
DPC value reaches its maximum (i.e. 10094). 
The main steps of the DPC calculation are presented in Fig. 3. 
  
  
Remotely 
sensed image 
Y Y 
Selection of 
control points 
Vector map 
  
  
  
  
Edge detection 
  
  
  
  
  
  
  
Detection of contour pixels 
with suitable direction 
Y 
Calculation of DPC 
  
  
  
  
  
Figure 3. Calculation of DPC.