If we suppose a successful selection of the representative 
primitives, the structural texture analysis detects the boundaries 
between two different textural objects with high resolution, 
contrary to the statistical texture analysis. On the other hand 
the human visual system responds on small statistic changes in a 
texture with high sensibility /7, 8/. Therefore a neglect of the 
statistical methods seems not to be justified and our investiga- 
tions are occupied with the combination of both methods. 
The detection of elements, which are defined by small spectral 
homogeneous regions, may be performed by the automatic 
multispectral classification, or if this Step fault, by spectral 
cluster analysis in preselected areas. 
To demonstrate the algorithm fig. 10. shows a result of a simple 
ciuster algorithm in the remaining areas. A unique spectral class 
with a typical composition of elements with different size and 
shape is represented in fig. 11. One subset of these elements 
with similar size and shape features is extracted and superposed 
with yellow colour into the original data in fig. 12. Relation 
of interest is the distance between each two neighboured elements. 
A "shrink and blow" algorithm applied here combines all elements 
and fuses them to a unique region, which is marked by a black 
border in fig. 12. This example demonstrates the detection of 
textured regions and of single objects too. 
|^» M&-114-91 ^ DONRU. BNC/€, 
* ter * i 
t Am 
7% 
= 
N 
  
+ 
ANALYSE: SPEKTRALE SIGNATUR. 
  
Fig. 10: Texture- Pig. 11: Class of Fig. 12: Detection of 
building elements Spectral similar Spatial clusters 
in remaining areas elements 
Processing all detected classes of elements a variety of different 
regions results. Some regions may spread over if the extracted 
elements do not represent the main primitives of the discrimina- 
tion problem. A postprocess may consist in fusing similar 
segments. To weighten the extracted segments their statistical 
texture features are now calculated in order to solve the fusing 
problem. 
PROPOSAL OF AN EVALUATION SYSTEM 
The proposed structure of an evaluation system is shown in fig.13. 
To facilitate the discussion of the system, the arrows between 
the different boxes, are assigned by numbers. 
40 
  
After | 
will be 
a) If ! 
res, 
as 1 
tior 
Sma] 
and 
Tf r 
set 
gene 
ture 
b) Tf 1 
ares 
rast 
imag 
feat 
of is 
seal 
shar 
prin 
elem