y using, interest
st points as the
aphs have been
e of the pairs of
digitized in the
sponding to the
ng, homogenous
There are 23
and 34 features
by using the
ors described in
neural network
ied and 10 pairs
The matched
2-17, 19-22, 20-
images visually,
matches.
iplate matching
iscovered. After
n, 16 conjugate
residuals of y
he mean square
square error of y
utation time was
in a Pentium 90
FI
&
Es
17
(d
3 40
b b 9,
5 S M
5 e,
1i
(sf 8
>
J
=
E
a
frm
IA
es
or
à t5
- d
ha
pa
fy À
[5 0.
4;
Ug
- ( Y
DER). ll | J ni
Ua (4, E C2 5 Lil
i ü
M
Figure 6: The boundaries of segmented homogenous areas from the test images
6. CONCLUSIONS
A fully automatic system is designed to orient digital
stereopairs of images. This system solves the initial
orientation by matching boundary features of
homogenous regions. The merit of matching features
is offering reliable initial orientation which provides
the possibility of finding accurate conjugate positions
by using template matching techniques.
For the matching of features, a neural network system
is implemented to match Fourier descriptors of features
with the consideration of shape similarity and
orientation consistency. This system works well even if
there are serveral similar features distributed in the
images. It is also adaptive to the disturbances of image
distortion and noises as well as the differences of
orientation and scale between images. We expect this
system will be useful in the fields of digital
photogrammetry, pattern recognition and computer
vision.
The sucessful experiments which we have tested on
several stereopairs of varying scale and ground
coverage encourage us to extend the system to orient a
block of photographs for automatic aerial triangulation.
The system should also be useful for the applications of
close-range photogrammetry. Future experiments will
be directed towards these two applications.
We expect that the method will work as long as the
boundaries of homogeneous regions in both images are
still similar in shape. Matched features should similar
in whole shape. If conjugate features are only partially
matched in shape, they will not be detected by the
system. However, this does not degrade the reliability
of the overall results.
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996
ACKNOWLEDGMENTS
This research project was sponsored by the National
Science Council of the Republic of China under the
grant No. NSC84-2211-E006-045.
REFERENCES
Dougherty, E. R. and Giardina, C. R, 1988.
Mathematical Methods for Artificial Intelligence and
Autonomous Systems, Prentice Hall Inc.
Freeman, J. A. and Skapura, D. M., 1992. Neural
Networks Algorithms, Applications, and Programming
Techniques, Addison Wesley.
Hopfield, J. J. and Tank, D. W., 1985. Neural
Computation of Decisions in Optimization Problems,
Biological Cybern, 52, pp. 141-154.
Lin, C. S. and Hwang, C. L., 1987. New Forms of
Shape Invariants from Elliptic Fourier Descriptors,
Pattern Recognition, Vol. 20, No 5, pp. 535-545.
Schenk, T., Li, J. C. and Toth, C., 1991. Towards an
Autonomous System for Orienting Digital Stereopairs,
Photogrammetry Engineering & Remote Sensing, Vol.
57, No 8, pp. 1057-1064. |
Tseng, Y. H. and Schenk T., 1992. A Least-Squares
Approach to Matching Lines with Fourier Descriptors,
International Archives of Photogrammetry and Remote
Sensing, Vol. 29, Part B2, Commission III, pp. 469-475.
Zahn, C. T. and Roskies, R. Z.,1972. Fourier
Descriptor for Plane Closed Curves, IEEE Trans. on
Computers, Vol. 21, No. 3, pp. 269-281.
