Roeland de Kok 
  
ANALYSIS OF IMAGE OBJECTS FROM VHR IMAGERY FOR FOREST GIS UPDATING IN THE 
BAVARIAN ALPS 
R. de Kok, A.Buck, T.Schneider, U.Ammer 
Lehrstuhl für Landnutzungsplanung und Naturschutz 
TUM Freising 
Roeland.dekok @Irz.uni-muenchen.de 
Working Group III/5 
KEY WORDS: Remote Sensing, Object-oriented, segmentation, semantic network, alpine forestry. 
ABSTRACT 
Pixel based analysis deals with the three basic features of a single pixel; Position Size and Value. The important spatial 
context of the pixel is used in filter operations, where the output value is assigned to the central pixel. In the middle of 
the 70", Landgrebe (1976) tried to register the important spatial context of the pixels (the software ECHO), assuming 
that a local pixel-group was representing a specific spectral distribution of the local land-cover class. Such analysis 
could only be successful when the problems with segmentation algorithms were solved. Cross and Mason (1988) 
continued this work, using quad tree segmentation techniques. In the work of Molenaar (1990), the basics of 
mathematics similarity between raster surface description and vector surface representations were shown. As this 
similarity was obvious, the powerful use of the SQL analysis of the database with raster (image) objects would be 
feasible. 
In the later work of Gorte (1998A), the importance of the quadtree segmentation image output in combination with a 
table link showed how GIS objects and raster (image) objects could be treated in the same way inside one database. 
Gorte pointed out the curious lack of attention in the standard literature on the potential of segmentation for remote 
sensing of earth observation. He also pointed out the pre-advantage of the remote sensing and GIS synergy which was 
considerably assisted with object based image analysis. The late 90 shows a remarkable increase of interest in the 
remote sensing community for advanced segmentation techniques and the possibilities of advanced SQL applications 
among image objects. In Munich the Delphi2 e-Cognition team has been concentrated since 1995 on the development 
of an image analysis software which combines an advanced segmentation technique with a database of image objects, 
which are linked through an hierarchical network. Their philosophy is based upon a Fractal Net Evolution” concept 
derived from the ideas of Dr. Binnig. Fractal Net Evolution is an efficient method for the description of complex 
semantics within largely self-organizing and dynamic networks. It combines insights into the fractal semantic of the 
world with object orientation. (Baatz, 1999). 
The forestry service in Bavaria has been aware that the new generation of satellite sensors could offer an alternative to 
the standard photogrammetric analysis of the state forest areas. With a predefined object , The Forest Stand’ and a 
standard forest map scale of 1:10.000, the imagery with 10 meter resolution up to 1 meter panchromatic data could offer 
enough detail to allow the extraction of important forest parameters from Very High Resolution data (VHR, 5 meter 
resolution and more). Previous studies from Kenneweg et.al. (1991) showed the difficulties of standard procedures of 
spectral analysis in forestry using VHR data from aerial platforms. Although successful in Landsat type of imagery, it 
became clear that the 1:10.000 map scale and the new generation of Satellites like Ikonos and Quickbird needed another 
approach The imagery could not only be used in visual interpretation, but the advantage of the digital data would be 
taken into account in an automatic analysis procedure. The utilization of the crucial spatial context of raster imagery 
were already shown in Neural network analysis and Wavelet transformations. Object based analysis opened up another 
way, were the filter size of the moving window defining the context became irrelevant and more important, allowed a 
construction of the database with spectral as well as spatial and topological features of the pixel population, making up 
the image objects. 
Experiments in the forest of the Bavarian Alps have shown, that automatic analysis is possible with 1 meter 
panchromatic data in combination with multispectral bands (deKok,1999). Important forest parameters, such as stand 
closure, deterioration, erosion and species composition can be derived automatically from such type of data in the 
difficult terrain of the Bavarian Alps. The database of Delphi2 eCognition merges image and GIS analysis, allowing any 
kind of imagery being integrated with the existing forest GIS. In this way making it possible to analyze this data, using 
fuzzy logic decision rules. The synergy of remote sensing and GIS, offers a good basis to assist the planning and 
decision process needed to maintain a stable mountainous forest so it can continue to play a multifunctional role in 
protection, production and recreation in the Bavarian Alps. 
  
222 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000.