TEEN 
Oniy the large compact regions are reliable results of the 
multispectral analysis. The other objects may be elements of 
texture regions. To distinguish between the different kinds of 
regions, the size and shape parameters may be an indication. 
Hitherto existing investigations show some weakness about the 
notion of compactness. "his fact is demonstrated in fig. 6, where 
a Set of size and shape parameters are applied to the data of 
fig. 4. Some small and less compact regions falsely are contained 
in the result. On the other hand sufficient compact regions have 
been eliminated. At last the structural texture analysis discussed 
later will distinguish these regions. To accelerate the algorithm, 
investigations are accomplished. 
d LULA 
WIENDI. KRE:  MIMADI-32 KREUZ-KL HALDATEN 
  
Fig. 4: Classification Fig. 5: Objects of Fig 6: Selected seg- 
one class ments of result 
STATISTICAL TEXTURE ANALYSIS IN REMAINING AREAS 
The statistical texture analysis supposes the definition of 
picture elements in order to calculate the texture features. 
Without a priori knowledge, it is usual to define regular raster 
distributed square elements. The texture parameters such as mean 
value, variance and moments of higher order from the histograms, 
grey level distribution along lines, run length statistics or 
two dimensional grey level dependences are extracted from the 
grey information in these elements /3/. Disadvantageous is that 
the texture parameters are only representative, if the size of 
the elements is adapted to the granulation of the texture. Near 
the border of an object the raster elements should be adapted 
to the coarse of the border. 
More effective is the evaluation of the statistical texture 
analysis only in the remaining areas, because the spectral 
homogeneous objects are already known by the preceding 
multispectral analysis. Additionally the boundaries between these 
objects and the remaining areas are defined. Fig. 7 shows the 
raster elements in the remaining areas to which some features 
‘Of the second order statistics of grey level dependence are 
applied. Elements with similar statistical features are fused 
into same classes and displayed with same pseudo-colour in fig.8. 
The obtained advantages are summarized as follows: 
38 
  
- "he € 
spect 
- Mixed 
adapt 
— Becau 
from 
class 
  
Tis. 7 
defini 
remain 
Tis. 9 
and th 
advant 
— Smal 
are 
— Boun 
text 
- Ifn 
only 
To avc 
some Ss 
1t.is 
primit 
In the 
must b 
of the 
of at 
Unfort 
number 
textur 
primit 
the inr 
Eom