Figure 6. Three representative, damaged buildings and the results with the proposed workflow. Grey areas with thin lines and transparent
areas with thick lines (figures (b) and (c)) correspond to classification classes
O
1*
and
O
2
, respectively. Figure (d) shows the result affected
by slivers after polygon intersection. Figure (e) illustrates the result after sliver removal.
the requirements of their specific applications. Then, the
spatial error in classification polygons is removed with the
proposed procedure for a rigorous, multitemporal comparison
of classification results in an efficient, object-based change
detection framework.
Future work consists of testing the proposed methodology
in different environments not only limited to urban areas.
Land-cover type can have a significant impact on object-based
classification algorithms, starting from the segmentation
phase to the choice of procedures for polygon comparison in
change detection. In fact, completely different segmentation
results are expected for natural environments (e.g., forested
environments) or agricultural areas with clear boundaries,
resulting in classification polygons with variable shape and
size. Future developments are related to the implementation
of strategies able to improve the preliminary registration step
based on piecewise affine functions. A denser set of corre-
sponding points would result in a better alignment of the clas-
sification polygons and a more reliable separation of spurious
slivers based on the geometry of elongated features.
References
Alexakis, D.D., E. Gryllakis, A. Koutroulis, A. Agapiou, K.
Themistocleous, I. Tsanis, S. Michaelides, S. Pashiardis, C.
Demetriou, K. Aristeidou, A. Retalis, F. Tymvios, and D.G.
Hadjimitsis, 2014. GIS and remote sensing techniques for the
assessment of land use changes impact on flood hydrology: The
case study of Yialias Basin in Cyprus,
Natural Hazards and
Earth System Sciences
, 14: 413–426.
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
February 2016
167
