corresponding to
OWA
average value was found to be ap-
propriate. This strategy proved to balance the effect of being
optimistic and pessimistic in constructing a complete road
network. It appears that being optimistic leads more gaps to
arise in the extracted roads, and thus network completeness
is reduced. Alternatively, selection of pessimistic strategy will
overestimate the extracted line segments which are not neces-
sarily correct. In none of these situations the accurate assess-
ment parameters are obtained.
One of the advantages of the proposed method is consider-
ing knowledge about the appearance of the roads with a small
number of geometric criteria, without using any radiometric
features, in the road key point connection stage. In this frame-
work, an innovative road geometric feature (
CI
) was proposed
to decide which road key points are inclined to be connected
to construct the true road network topology.
The proposed methodology is capable of vectorizing
roads in presence of small disturbances on the road surface.
Besides, this methodology is efficient in the case of roads with
different widths, closed shapes in the road network, and also
nearby parallel roads comparing with
SORM
and
IEC
and fuzzy
clustering method. These advantages improve the quality
measures prominently. Another achievement of the research
is the robustness of the proposed algorithm against noise clus-
ters. The robustness of the algorithm in presence of small non-
road objects decreases the falsely extracted road segments.
By the way, it is obvious that the outcome of the vectoriza-
tion stage is highly affected by the road detection output. The
better is the result of the detection stage, the more complete
and accurate vectorized road network is attained. In this re-
search different road detection techniques were employed. The
evaluation results justify the robustness of the proposed vector-
ization approach applying different detection methodologies.
Another important aspect is using a few parameters includ-
ing K, R, and T
cost
by the algorithm. The parameters R and T
cost
are quite stable, as shown by the fact that they were tested on
multiple images from different scenes and different sensors. It
is necessary to select the radius of the overlapping circles (R),
higher than half of the distance between two road key points. In
all the experiments parameter R between [0.6d , 0.7d] was ap-
peared to be appropriate. R >0.7d results in wrong connection
especially in place of nearby parallel roads. Values in the range
of [0.5d , 0.6d] reduce
CI
, which leads to higher cost values
and missed road segments. In selection of the cost threshold
value, If
d
i
≤
CI
i
, then the cost will be in the range of [0,1]. On
the other hand, if
d
i
>
CI
i
, then the cost will become larger than
“0” (Cost
≥
0). Eventually, the value of T
cost
was experimentally
defined as “1” to ensure the most complete and accurate road
network. Only one parameter (K) had to be changed regarding
the types of intersections in the images. Although, selecting the
same maximum number of K for all the images does not have a
considerable effect on the results; selecting smaller values only
restricts the searching space for the road key points around.
However, our experimental results indicate a number of
limitations. It is often not possible to extract a whole road
completely in the case of disturbances in the appearance of
roads that results in a large gaps in the detected road. Another
deficiency of the proposed connection method is construction
of enclosed area at place of junction where the width of con-
necting roads is high. To overcome this disadvantage a precise
intersection modeling should be considered.
Conclusions
This paper proposed and verified a novel road vectorization
methodology using image space clustering technique and
weighted graph theory. According to the experimental results, the
proposed road vectorization method is efficient and accurate in
road network topology construction at different types of inter-
sections, roads with dissimilar widths, parallel roads, and also
images containing isolated non-road clusters. Furthermore, it
shows robustness in vectorizing a road network regardless of the
detection methodology. This method has also enough generality
to include appropriate connection criteria to vectorize other types
of linear cartographic features. The results are acceptable enough
to be used in
GIS
databases with negligible operator interferences.
Precise modeling of the context objects and intersections can im-
prove the topological and geometrical quality of the results, which
can be considered as the primary direction for the future research.
Acknowledgment
The authors would like to gratefully acknowledge the many
valuable suggestions made by Najmeh Neisani Samani, assis-
tant professor at Tehran University.
References
Ahn, B.S., 2008. Preference relation approach for obtaining OWA
operators weights,
International Journal of Approximate
Reasoning
, 4:166–178.
Ameri, F., A.M. Mobaraki, and M.J. Valadanzoej, 2008. Semi-
automatic extraction of different-shaped road centerlines from
MS and pan-sharped Ikonos images,
International Archives
of Photogrammetry, Remote Sensing and Spatial Information
Sciences
, Vol. XXXVII, Part B3b, Beijing, China..
Amin, G.R., and A. Emrouznejad, 2006. An extended minimax
disparity to determine the OWA operator weights,
Computers &
Industrial Engineering
, 50:312–316.
Bacher, U., and H. Mayer, 2005. Automatic road extraction from
multispectral high resolution satellite images,
CMRT05
,
International Archives of Photogrammetry and Remote Sensing
,
Vol. XXXVI, Part 3/W24.
Baumgartner, A., C. Steger, H. Mayer, W. Eckstein, and H. Ebner,
1999. Automatic road extraction based on multi-scale, grouping
and context,
Photogrammetric Engineering & Remote
Sensing
65:777–786.
Chaudhuri, D., N.K. Kushwaha, and A. Samal, 2012. Semi-
automated road detection from high resolution satellite images
by directional morphological enhancement and segmentation
techniques,
IEEE Journal of Selected Topics in Applied Earth
Observation and Remote Sensing
, Vol. 5.
Chen, L.H., C.C. Hung, and C.C. Tu, 2012. Considering the decision
maker’s attitudinal character to solve multi-criteria decision-
making problems in an intuitionistic fuzzy environment,
Knowledge-Based Systems
, (36)129–138.
Cornelis, C., N. Verbiest, and J. Jensen, 2010. Ordered weighted
average based fuzzy rough sets,
Proceedings of the 5
th
International Conference on Rough Sets and Knowledge
Technology (RSKT 2010)
, pp. 78–85.
Cover, T.M., and P.E. Hart, 1967. Nearest neighbor pattern
classification,
IEEE Transactions on Information Theory
,
13:21–27.
Doucette, P., P. Agouris, A. Stefanidis, and M. Musavi, 2001. Self-
organised clustering for road extraction in classified imagery,
ISPRS
Journal of Photogrammetry and Remote Sensing
, 55:347-358.
Emrouznejad, A., 2008. MP-OWA: The most preferred OWA operator,
Knowledge-Based Systems
, 21:847–851.
Ferchichi, S., and S. Wang, 2005. Optimization of cluster coverage for
road center-line extraction in high resolution satellite images,
Proceedings of the IEEE International Conference on Image
Processing
, pp. 201–204
Fuller, R., and P. Majlender, 2001. An analytic approach for obtaining
maximal entropy OWA operator weights,
Fuzzy Sets and
Systems
, 124:53-57.
Garber, N., and L. Hoel, 2010.
Traffic & Highway Engineering
, Cengage
Learning Publishing, Fourth edition, ISBN:0495438537.
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
February 2016
119
