20 
dicative of the incomplete system training* 
During the segmentation of the image into 
the indicated classes all the pixels are 
painted in colours corresponding to the ide 
ntified classes* After that automatic conto 
uring of regions with similar colours is do 
ne* The obtained contour network is the so 
urce material for the further processing 
when the thematic map of the land categori 
es is compiled* 
The application of the mentioned method 
to space scanner photos makes it possible 
to get the smoothed contours of the extrac 
ted plots* If photographs obtained with the 
MKF-6 camera are used, then as a result of 
the more complicated and delicate texture 
of the image the contours are very broken, 
and the plots contain a lot of small inser 
tions* The image becomes porous* To elimina 
te pores the image texture is analysed and 
the elements' area is determined* The thre 
shold value of the area below which the plo 
ts are considered outside insertions and must 
be eliminated is established* The elimina 
tion is done in the following way* 
Each small plot is analysed element by e- 
lement along the boundaries, and the element 
class of the area adjacent to it is determi 
ned* The pixel of the eliminated plot is 
related to this area* In such a way all the 
pixels of the eliminated plot, element by 
element, are passed over to the neighbouring 
areas* After the process is over all the co 
ntoured plots have the areas that are grea 
ter than the threshold and do not contain 
insertions* 
Of great interest is the estimate of the 
correctness of going along the selected plo 
ts contours* The difficulty of such an esti 
mate lies in the fact that the real signifi 
cance of areas and true contours are as a 
rule unknown* That is why one has to resort 
to indirect estimates* If we consider the 
computer segmentation as a model of visual 
segmentation done by the interpreter in the 
course of contour photointerpretation, the 
indirect assessment of the correctness of 
the machine segmentation may be considered 
as the adequacy of its visual segmentation* 
The adequacy test was carried out by the 
example of selecting contours of one and the 
same stand class (pine stands) on the colour 
synthesized MKF-6 space photograph* In this 
test 10 specialists experienced in visual 
interpretation of space photo6 did the pine 
stands contouring independently* The conto 
urs were entered into the display system* 
The statistics of the variations between the 
visual interpretation results characterizing 
the unit were shown in the following way* 
The image of the contoured region was consi 
dered as a multitude of Q pixels equal in 
number to the region's conventional area* 
The contours deviation of the i-th and j-th 
interpreters was determined by the formula: 
ku^WL5i!., 00 ¡¿j 
I dii 
whereU,0 are the symbols of the multitudes 
unification and intersection* 
The experiment consisted of ten series by 
which selective deviation dispersions cha 
racterizing the differences between the vi 
sual interpretation results were calculated* 
By the Kochran criterion the hypothesis of 
the series homogeneity is confirmed* Then 
the dispersion of the machine contours devi 
ations from all visual contours was calcula 
ted, and the adequacy of the machine and vi 
sual segmentation was established by the 
dispersion ratio* 
Another method of the contour-line extrac 
tion on photos is used in case when a part 
of lines is not depicted visually on photos* 
These are administrative boundaries of plots 
which do not coincide with natural borders, 
boundaries of forest compartments differing 
by mean values of the forest mensuration 
indices: tree height, density, age, etc*, 
different types or isolines: relief, lines 
of equal reservoir depth, isotherms, etc* 
Such is the case when forest management map- 
cases are compiled by photos* For this pur 
pose the method of the entering of contour 
lines from the photo outline has been deve 
loped* 
All contour lines are first transferred 
by the interpreter to the photograph in whi 
te gouache in order to make up a photo out 
line* Then the information from the photo 
graph is automatically entered into the sys 
tem together with the photo outline conto 
urs* Tnese contour-lines extraction is done 
with a special program package which reali 
zes the method of the 'nonsharp* and 'logi 
cal* masks* 
In the 'nonsharp' mask method the positive 
component of the difference between sharp 
and nonsharp images is formed* It contains 
more complete information about contour li 
nes* The differential image multiplied by 
the coefficient K is added to the source 
image* The result is shown on the display 
screen and may be corrected by means of cha 
nging the value of K* The 'nonsharp' mask 
favours the tone smoothing, and contour line 
underlying and thinning* 
The obtained image is further processed 
by the 'logical' mask* It is a scanning win 
dow of 3x3 in which the comparison of brigh 
tness of the central pixel and pixels of the 
neighbourhood with the level P and with each 
other in accordance with the set logical 
condition is done* When meeting the conditi 
on the central element is marked which cor 
responds to the extraction of the contour 
line point* 
2 CONTOUR LINES RESTORATION 
It is an important stage of the contour net 
work processing after its extraction* It is 
meant for the elimination of the residual 
defects in lines: breaks, thickening, thin 
ning, unnecessary clusters, etc* The proces 
sing is done on the basis of the analysis of 
the image point neighbourhoods* The neighbo 
urhood contents are indicative of the conto 
ur defects class in the central point, that 
is why on its basis the decision about the 
marker sending and cleaning in the central 
cell is taken* For each defects class defi 
nite image points are set* Their neighbour 
hoods must be analysed no matter marked or 
unmarked they are* For example, to eliminate 
breaks the neighbourhoods of cells in which 
the contour (marker) is absent are analysed* 
We shall call digital image cells in which 
the neighbourhood is analysed the 'current 
cells'* Let us consider the quite broad nei 
ghbourhood of the current cell M * We shall 
single out in it square ring-shaped zones 
around M of 3x3, 5x5, 7x7, etc* The side of 
the n-thzone equals to 2n+l elements of the 
image* The total number of elements in the 
n-th ring-shaped zone equals to (2n+l)4-4» 
“8n* As eight binary elements make up one 
byte, any n-th zone may be shown in a code 
correspond! 
Let us d< 
around the 
develop it 
te the cont 
of the neig 
byte into > 
marking th< 
turn lines v 
each elemer 
the Zj-neig 
By means 
neighbourhc 
into class« 
classes frc 
classes nan 
class A she 
ends of the 
ghbourhood 
tion of dif 
bourhood st 
Each class 
rule of the 
be sent to 
sent to M * 
Each rest 
following m 
search, ana 
ch current 
and determi 
Y-half-byte 
the state ra 
and fulfilm 
ansition to 
nage to sin 
tour defect 
3 CONTOUR N1 
The binary i 
system is t< 
it can be n 
the submers: 
base and fo: 
made to the 
lines must 1 
pixel of th< 
xels of the 
their sides, 
closed circt 
During th< 
of the cont< 
the vector i 
le reunified 
ed sections 
threads dep< 
and positioi 
rectificati« 
revision poi 
the map, anc 
carried out, 
In order t 
works the di 
developed (E 
so called 't 
which allows 
along the 11 
direction. I 
necessary pi 
The essenc 
lowing* Abov 
the marker n 
inquiry prog 
tal probe) i 
ved by the p 
the element 
jacent eleme 
ur direction 
the left, up