erations 
îtion 
CPU time 
(min.) 
5 
5 
3 
15 
150 
125 
105 
85 
65 
45 
30 
15 
Symposium on Remote Sensing for Resources Development and Environmental Management / Enschede /August 1986 
Methods of contour-line processing of photographs 
for automated forest mapping 
R.I.Elman 
All-Union ’Lesprojekt’ Association, Moscow, USSR 
:eded should be the 
according to the 
s no siginficant 
,B*R. But AV8B*R spent 
therefore C8B-R 
ABSTRACT: Automated extraction of forest boundaries on aerial and space photographs is con 
sidered» The methods depend on the contour line clearness on photos and differ in the auto 
mation degree. The contour extraction is followed by restoration aimed at elimination of de 
fects in contour lines which appeared during the input» The next stage is the random conto 
ur networks processing» It is meant for the transformation by contour lines from the pattern 
into the vector format in order to cut the network into elementary unraraified ’threads’, to 
code threads, etc» The method is realized in the form of the program package for the disp 
lay processor and is used in the forest mapping automation. 
find the optimal 
S data for forest 
following conclusions 
own that the past 
cover type classifi- 
all wavelength bands 
, one can find that 
ns more bands, 
ring techniques 
curacy because both 
pening the images to 
) Principal component 
n improving the 
lassification because 
ast enhancement for 
e also not good enough 
ation because the 
e difference in 
s with different 
tio images. (5) Six 
bands are unable to 
The mixed bands 
s their appreciation 
ommittee, Council of 
for providing the 
so due to the Center 
sarch, National 
cessing services. 
1.A. Mead 1983. 
)n accuracy using 
statistical tech 
iring and Remote 
L983, pp.1671-1678. 
Computer-aided 
rational system to 
isat MSS data, LARS 
» University, West 
L-231. 
Accuracy of forest 
Compatible tape, 
itional symposium on 
PP.1155-1163. 
; species mapping 
sing of Environment 
1 CONTOUR LINES EXTRACTION 
In the forest mapping automation with the 
use of Information obtained from aerial and 
space photos one has to do with operations 
on contour networks formation making up the 
contour basis of the complied map» These 
operations may be quite different by the 
characteristics, labour-consuming nature and 
complicacy depending on the primary method 
of contour lines obtaining» 
In the world practice the semi-automatic 
Input of the graphical information with the 
help of digitizers Is widely used» Such in 
put Is effective when the contour network is 
not too complex and has a lot of rectilinear 
segments» The preference is given to it also 
In case of superimposing several heterogene 
ous contour networks, the automatic separa 
tion of which is difficult» However this me 
thod is slow and axhaustlng as it requires 
a lot of manual work» 
A higher degree of automation and accelera 
tion of the process of contour lines input 
can be achieved by using television or opti 
cal-mechanical input devices» However in this 
case means of input and extraction of conto 
urs on photographs depend on the kind of 
units to be delineated» For example, units 
having a clear, high contrast image on aeri 
al and space photos are successfully deline 
ated automatically, with the help of simple 
threshold procedures or filters described in 
detail in literature (Prett 1982) (Roberts, 
Sobel, Kirsch et al»)» Such units in silvi 
cultural tasks are: forest boundaries, fresh 
burns, cutovers, etc» But still the major 
part of the necessary contour lines is left 
unextracted» That is why this method is used 
rarely, in particular model tasks» 
A more complicated but more effective me 
thod of the extraction of contour lines se 
parating different natural units from each 
other is based on the image segmentation 
with a further contouring of the extracted 
segments» In our work (Elman, Bakhtinova, 
Potapov, Sviridova & Bersneva 1984; Elman & 
Pamorozski 1986) colour and partially textu 
re features were used for image segmentati 
on» For example, in (Elman, Bakhtinova, Po 
tapov, Sviridova & Bersneva 1984) the seg 
mentation of space scanner photos obtained 
with the ’Fragment* equipment (Selivanov, 
Gektin, Panfilov A Fokin 1981) was carried 
out» In experimental work video information 
written on the magnetic tape, in three spec 
tral bands (0»6-0»7, 0»7-0»8 and 0»8-l»l mic 
rometres) was used» 
According to the representation of land 
categories, economic significance, and the 
r ossibility of the test plot control the fol- 
owing set of classes was determined: A - 
pine stands, B - birch stands, C - aspen sta 
nds, D - water surfaces, E - land non-covered 
with forest (arable land)» The test plots of 
all classes were chosen by the forest stands 
plans at scale 1:25000 with further field 
investigations» The plots were selected in 
side homogeneous compartments with the area 
of 20-30 ha» Minimal plots of the fixed area 
of about 5 ha were taken as control plots 
by which the experiment results assessment 
was made (Table 1)» 
Table 1» The results of desses extraction 
by colour features» 
K 
Machine 
solutions 
Q 
N 
P 
A B 
C 
D 
E 
number 
7. 
A 
63 
63 
63 
100 
B 
50 
13 
63 
50 
79 
C 
14 
33 
07 
54 
33 
61 
D 
54 
54 
54 
100 
E 
92 
07 
99 
92 
93 
Mean 
estimate 
~rrr 
292 
~S7~ 
Notes: K - actual classes 
Q - number of omissions 
N - total plots 
P - identified correctly 
The table analysis shows that classes of pine 
stands and water surfaces are Identified with 
the best reliability, and birch and aspen 
stands are mostly confused» There are omis 
sions in the arable land class 'which is in-