32 
'1!; so the areas of the node will become very extensive and 
hence not well definable. 
It has to be noted that sometimes, owing to the difference in 
surface light density of che various parts of the raster,the use 
of one single level of threshold for the whole matrix A(i,j) may 
cause a change in the information content of the image, producing 
in this way geometric distortions and consequent loss of accuracy 
in the subsequent phases of processing. This can be avoided by 
using image restauration techniques before the thresholding pro- 
cedure, such as contrast enhancement and/or edge enhancement, 
¢} Thinning 
This procedure (8) aims at reducing the thickness of every line 
to cone single pixel, replacing practically every stripe produced 
by the thresholding procedure with its central line. A new 
"thinned" matrix A"(i,j) isset up (Table 4). 
Table 4 
"Thinned" Image 
  
In the ideal case the node should be made up of the only pixel 
of intersection of one horizontal and one vertical line, and it 
is hence easily identifiable by analysing the 8 adjacent pixels. 
However, generally in practice, if the deformation of the raster 
is great, the nodes will still be constituted by an area of pi- 
xels of value "1°, 
Each of these pixels is, in any case, well characterized by the 
fact that the sum of the values of its 8 adjacent pixels is 
always 23. The node point will be represented by the barycenter 
of this node area. 
me 
t" 3 
TI 
asurement of coordinate 
gure 2 shows the thre