Pattern recognition after preprocessing 
If texture is assumed to be inherent in local grey level depen- 
dencies, it is also possible to describe the texture informa- 
| tion of the grey level pictures by evaluating their derivatives. 
| In this case, the two dimensional grey level image will be 
transformed into a matrix of the same dimension and size, where 
the elements represent local contour information. There are a 
lot of approved differentiation methods /3, h/ to generate 
gradient images with directional and weight information. Fig. 5a 
shows the original data and 5b the corresponding gradient image, 
where the absolute weight values of the gradients are displayed 
as grey level intensities. To extract textural features from 
the absolute values the same histogram analysis described for 
the grey level images can be used, Computing the weight distri- 
bution instead of the grey levels we have a stronger connection 
to the texture. For statistics about the directions within an 
evaluation unit some modifications of the histogram analysis 
are needed. However, the patterns taken from histograms of di- 
rection distributions do not seem to be as good as those from 
the absolute values of the gradients. 
To include the direction information and to consider more global 
relationships it is preferable to apply a further preprocessing 
procedure /3/. This procedure extracts from the gradient infor- 
mation line drawings or line lists from which a lot of textural 
patterns can be derived. Some of those patterns are e.g. the 
number of lines within the evaluation unit, the mean value and 
variance of the line length distribution, the ratio between 
the straight lines and the total of lines, the degree of parallel- 
ism of the lines, and the angular distribution at the points of 
line intersections.