typical example of obtained results.
Fig. 6: A portion of a SPOT panchromatic image, with
marked seed points and potential problematic areas
Fig. 7: Wavelet transformed image
Specifically, Fig. 6 depicts a portion of a SPOT pan-
chromatic image of Moudon, Switzerland. On it,
marked by black discs, one can distinguish the seed
points provided by the operator along the road which
is to be extracted. The white rectangles denote po-
tential problematic areas, i.e. intersections with other
roads which might cause an erroneous change of di-
rection, and increased radiometric noise and ambi-
guities due to passage through an urban area. Fig. 7
shows the wavelet transformed image. It is apparent
that the transformation has made the existing road
network appear more prominently, thus facilitating
the subsequent extraction of the road of interest, the
result of which is shown in Fig. 8. It can be easily ob-
served that no problems were finally encountered in
its extraction despite the presence of the aforemen-
tioned problematic areas. Additional experiments
with other types of potential problems (e.g. gaps, low
texture) verified the efficiency and effectiveness of
the technique.
Fig. 8: The semi-automatically extracted road
We have also successfully implemented least
squares template matching and Fig. 9 presents at
left a typical example of edge tracking by this method.
Matching has been initiated at the top of the white
dotted line and proceeded downwards. White dots
denote the resulting edge positions. Edge tracking
Stopped at the bottom of the dotted line (position
marked by white circle) because it reached a corner,
and as previously discussed this caused matching to
fail and tracking to stop.
Fig. 9: Least squares template matching results
The right side of Fig. 9 shows the application of this
method for the extraction of several edge segments
of an object and demonstrates cases of distraction or
152
failure due t
A comparis:
active conto
Fig. 10 shov
As expecte
fered a larg
and can be
tracted are |
contour moc
nature enab
ometric or €
Fig. 10: Ec
(left) and |
Globally ent
presented ii
ment, is ex]
of both tech
The semi-ai
sented in thi
ting on a c
have their :
make each
tensive vari
applying the
DTMs, surv
signed to th
highly auto
capture sch
improve the
some and r
Ballard D.H
Prentice He
Baltsavias