ellipse-shaped clustering technique on the three data sets are
presented in Figure 15a through 15c, 15d through 15f, and
Figure 15g through 15i, respectively.
The results of the evaluation are summarized in Table 6.
The evaluation shows that the vectorization results of different
approaches are almost similar with respect to sub-pixel
RMS
E
values. On the other hand, the proposed methodology is able
to extract a larger road network and to achieve a higher com-
pleteness values in comparison with the other methods. The
first reason is the success of the proposed centerline extraction
method in presence of missing information on the detected
roads. The second reason is that the proposed method is able
to vectorize closed shapes in the road network (depicted by
dotted rectangles in Figure 15a and 15b). In contrast, since the
SORM
uses the
MST
algorithm in the node linking stage, it has
some deficiencies in extracting a complete road network.
Concerning the junction
RMS
E in Table 6, the
IEC
and fuzzy
clustering method outperforms the other two approaches
at road junctions; they are able to extract the road junc-
tion’s position by means of the controlling fuzzy parameters
(Mokhtarzade
et al
., 2010). Falsely extracted junctions by the
proposed method most frequently happened where the width
of connecting roads are high at the junction’s intersection
area. In this situation some nearby road key points exist at the
intersection which forms an enclosed area (shown by dot-
ted circles in Figure 15c). Generally, the performance of the
proposed method is superior to others in both completeness
and correctness.
Discussion
The main objective of this research is to investigate the
ability of
OWA
based aggregation strategy to solve a problem
with a unique solution (a true road network). In this context,
different decision strategies between being extremely pessi-
mistic and extremely optimistic were tested and the strategy
(a)
(b)
(c)
(d)
Figure 13. Results of an Ikonos image of Shiraz: (a) original image, (b) detected road image, (c) result of the proposed methodology, and
(d) results of MST.
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
February 2016
117
