International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B4. Istanbul 2004 
The process can be modeled in a flow chart or script language intuitive and iterative approach which does not facilitate 
environment and thus be applied to other images and automated processing (Schiewe and Ehlers 2004). 
geographic regions. It has to be noted that the choice of suitable 
feature classes is still an interactive process and has to be Image indices, on the other hand, can be generated by simple 
performed by the user. procedures and are easily implemented. To a large degree, they 
can also be automated. They permit the generation of a number 
of meaningful feature classes if they represent a normalized 
process. Consequently, we chose the standard normalized 
difference vegetation index (NDVI) because it allows the easy 
separation of vegetated, non-vegetated, and water areas. The 
NDVI is calculated as the ratio of the difference between the 
near infrared and the red band and the sum of the two bands: 
NDVI = (nir-red)/(nir+red) 
Using this index, the difference between vegetation and non- 
vegetation is emphasized. Reflectance values for vegetation 
have their maximum in the near infrared and a minimum in the 
red spectral domain. High values of the NDVI indicate lush 
vegetation, values around 0 non-vegetated land areas and 
negative values are usually associated with water. Figures 8 and 
9 show the NDVI as gray value display and its histogram for 
the study site. For this investigation, we used a Quickbird image 
of the same area. 
  
Figure 6. GIS based enhancement of the IKonos image (for 
comparison see Figure 2) 
  
Figure 7. The subset of the GIS enhanced image (left) shows a 
higher level of detail compared to the globally enhanced image 
(right) (contrast stretch with +20) 
This process can be augmented using image segmentation 
i Image d for Ra KR em ghe Figure 8. NDVI for the Quickbird image of the study site. 
be used if a priori information is not available, outdated, of Bright values indicate high level of vegetation, intermediate 
inaccurate. values open fields, and dark values water 
3.2 Context Based Image Enhancement war © 27:06 0,743304 
Bub 
Often, there is not enough a priori information for a GIS based 
feature selection process or the information is not accurate 
and/or outdated. In this case, the information contained in the 
image itself is the most reliable source for feature based 
enhancement. The image has to be segmented into meaningful Comyn] 
features classes which are again mutually exclusive and can be 
used as masks to create independent image layers. The process 1080 
should have the potential of being formalized and automated so 
that it can run as a preprocessing step. There has been 
significant progress in the generation of meaningful segments 
for image classification, most notably the Fractal Net Evolution 
3340 
  
  
Approach in the software package eCognition (Blaschke and Figure 9. Histogram of the NDVI image. The separations 
Strobl 2001). This procedure, however, involves a rather between potential classes (local minima) are clearly visible. 
400 
  
  
Internatio 
ied dite 
The only 
selection 
procedure 
on local 
process s« 
on top of 
be interac 
been veri 
independe 
process b 
Table | p 
classes. I 
result of 
the same 
water are 
for better 
| 
  
  
  
  
For a bet 
shown ir 
superiori