iscussed 
gorithm, 
* 
d seg- 
f result 
S. 
raster 
S mean 
grams, 
the 
that 
e of 
Near 
ted 
n these 
the 
res 
sed 
fig.8. 
  
E 
" 
  
  
- "he exact detection of the boundaries between textural and 
spectral homogeneous objects is ensured. 
- Mixed textural features of the elements are avoided by the shape 
adaption between the elements and the object boundaries. 
- Because of the absence of the additional features extracting 
from the homogeneous objects the separation between textural 
classes becomes better. 
  
  
  
  
  
  
  
  
  
  
  
  
  
Va T T 
  
Combination 
of results 
Pig. 3: Result of 
statistical texture 
analysis 
Pia. 9: 
7: Raster- 
definition in 
remaining areas 
Fig. 
Fig. 9 shows a combination between the result of texture analysis 
and the result of multispectral analysis. It demonstrates the 
advantages mentioned before. Disadvantages are: 
— Small single objects and areas without statistical relevance 
are not processed by the statistic texture analysis. 
- Boundaries between two neighboured regions with different 
textures are detected with low resolution. 
- If no extraction of homogeneous spectral objects is succeeded 
only large homogeneous textured of areas are extracted. 
STRUCTURAL TEXTURE ANALYSIS IN THE REMAINING AREAS 
To avoid the specific weakness of the statistical texture analysis 
some structural texture analysis methods have been investigated. 
It is assumed that & textural object is compoesed by a number of 
primitives with characteristic features and spatial relations /5/. 
In the first step the primitives which compose the texture regions 
must be described. The aim of the second step is the description 
of the spatial dependence or interaction between the primitives 
of a texture. 
Unfortunately, many real image textures are composed by a large 
number of different elements. Only some of them already are 
texture primitives. In most cases only a subset of these 
primitives is able to distinguish between different textures in 
the image /6/. 
39 
rE ee