5 b) Rule5 c) Rule 6a
tt
d) Rule 6b e) Rule 7,6a
LP LP
g) Rule 7,6a h)Rule7,6a,6b,76a i) Rule 89,8,9,8
uf RÉ
j) Rule 10b k) Rule 11,6a,6b,7,6a 1) Rule 8,9,8,9,8
5
f) Rule 6b
D
Figure 3. Line parser run on synthetic segmentation.
6 EXAMPLE
The line segment parser was tested manually on a
synthetic segmentation of a building shown in figure
3a. The different rules used for the reconstruction are
shown step by step in figures 3b-l. As is easily seen,
rule 6b erroneously introduces vertical lines twice
(figures 3d and 3k), this due to sloping roof lines
directed towards the nadir point. These errors are
corrected in the consistency check, requiring that
segments be closed and plane. Further discussion of
examples are given in (Gülch, 1992).
7 DISCUSSION
The procedure described here splits the problem of
interpretation into two steps: segmentation and
parsing. For segmentation, long being a bottle-neck of
image analysis, procedures now exist giving results of
acceptable quality. However, the requirement in this
paper that segmentation boundaries should be given
by the same models as those used by the parsers, is
crucial: without this requirement parsing instead of
segmentation becomes an approximation process.
Parsing, being a task only involving logical decisions,
has nothing to do with approximation. Segmentation
procedures with this property are now being de-
veloped. Our implementation has not yet proceeded
far enough to give inputs forthe parsers described.
734
When this goal is achieved, more realistic tests will be
performed with the parsers presented here, which
have given very promising interpretations when
using synthetic inputs.
8 ACKNOWLEDGEMENTS
The support of the Swedish National Board for
Technical Developement under project nr 89-1537 and
the Swedish Council for Building Research under
project nr 890743-2 is gratefully acknowledged.
9 REFERENCES
Ballard, D. H., 1981. Strip Trees: A hierarchical
representation for curves, Comm. ACM 24, 310-321.
Dengler, J., 1991. Estimation of discontinous
displacement vector fields with the minimum
description length criterion. Proc. CVPR^91,
Lahaina, Maui, Hawaii, IEEE Press.
Dickinson S. J., Pentland A. P., Rosenfeld A., 1990.
Qualitative 3-D Shape Recovery using Distributed
Aspect Matching. CAR-TR-505, CS-TR-2484,
AFOSR-86-0092, University of Maryland, College
Park, Maryland.
Fua P., Hanson A. J., 1988. Generic feature extraction
using probability-based objective functions. SRI
International.
Fórstner, W., 1988. Model based detection and location
of houses as topographical control points in digital
images. In: Int. Arch. Photogramm. Remote
Sensing., Kyoto, Vol 27, B11, pp. 505-17.
Fórstner, W., 1989. Image analysis techniques for
digital photogrammetry. Proc. 42:nd
Photogrammetric Week, Stuttgart, FRG.
Georgeff, M. P., Wallace, C. S., 1984. A general
selection criterion for inductive inference. In
ECAI 84: Advances in Artificial Intelligence.
Elsevier, pp. 473-482.
Gülch, E., 1992. A knowledge based approach to recon-
struct buildings in digital aerial imagery. In: Procee-
dings XVII ISPRS Congress, Washington D.C. USA.
Herman, M., Kanade, T., Kuroe, S., 1984. Incremental
aquisition of a three-dimensional scene model
from images. IEEE Trans. Pattern Analysis and
Machine Intelligence, Vol. PAMI-6, No. 3.
Leclerc, Y. G., 1989. Constructing Simple Stable De-
scriptions for Image Partitioning. IJCV, 3, pp.73-102.
McKeown, D. M., 1991. Information fusion in
cartographic feature extraction from aerial imagery.
In Ebener, H., (ed.) Dig. Photogramm. Systems.
Karlsruhe: Wichmann.
Rissanen, Y., 1989. Stochastic Complexity in statistical
inquiry. World Scientific, Teaneck, NJ.
Walz D. L., 1972. Generating Semantic Descriptions
from Drawings of Scenes with Shadows. MIT,
Internal report. Reprinted as "Understanding Line
Drawings of Scenes with Shadows" in Winston
P.H. (Ed.) (1975): The Psychology of Computer
Vision, MacGraw-Hill, pp 19-92.
Zielinski, H., 1992. Line photogrammetry with
multiple images. In: Proceedings XVII ISPRS
Congress, Washington D.C., USA.
