To avoid hazards and get more statistical reliability the evaluation
unit should be covered with multitude of lines either parallel or
in different direction.
Tuoidinensjonalesparial relationship
A further simple procedure to extract textural patterns out of
grey level images is given by the computation of grey level de-
pendencies between the pixels and their surrounding neighbours.
Haralick e.g. has based such evaluations on the statistics of
the bilateral grey level dependencies between pairs of points
with a fixed distance and angular relationship (see fis. ^).
For each distance d and angle « the frequency of cooccurence
of the grey value i and j within the evaluation unit will be
computed /2/. This results in spatial dependent grey tone matri-
ces CLE which can be merged into mean matrices with con-
stant d and varying o. or vice versa. The statistical analysis
of the hax or the mean matrices yields patterns to describe
textural phenomena like homogeneity, contrast and structure
(ean.(3)). But often it is not so easy to find the correlations
between the patterns and the texture recognized by visuell inter-
pretation. In addition it is inappropriate to make use of all
patterns derivable from the matrices because for many classes
of objects some of these features prove to be highly correlated.
^ 2 M
KT = z G1) hing)
^
Has X46
1,3
(5)
EM -- Ling) Log (Lin)
^
: yg (AM) G- MG: Á t4) fea) C G)
tJ
with M., C
KR
2 according eqn, (1)