Design and Implementation of SPOR for GEOBIA
In this section, the design and the implementation of
SPOR
for
GEOBIA
are presented. To gain computation performance,
SPOR
was developed in C++. For multilevel segmentation analysis
(Baatz and Sch pe, 2000; Tzotsos and Argialas, 2006), two
topologic relationships were considered to represent the
hierarchical and spatial relationships of the segments. Each
segment requires knowledge of its neighbors at the same level,
its sub-objects at lower levels and its super objects at higher
levels. Before the reasoning process, if these relationships
have not been computed by the segmentation algorithm,
SPOR
carries out the required computations through SQL queries.
During the reasoning process,
SPOR
iterates over the classes.
For each class, the membership value for each segment (
MV
;
Figure 2) is determined. As
OWL
2 semantics do not provide
information regarding the depth of the hierarchy of the exam-
ined class, the reasoner was designed to iterate over the parent
classes of the examined class (Figure 2). If the membership val-
ues of the segments for the parent classes have not as yet been
computed, then the reasoner tries to compute their membership
values. To consider objects with adequate membership value for
all parent classes, as candidate segments for the current class are
examined those with membership value greater than 0.5 for all
parent classes. If the candidate segments of a class have a spatial
relationship with segments of other classes, the membership
values of the candidate objects with these classes are computed.
Afterwards the spatial relationships are computed (Figure 2).
For each candidate segment, a membership value is com-
puted. If the membership value of the candidate segment is
greater than 0.5, then the class name of the currently exam-
ined class is assigned as label to the examined candidate.
Finally the results are stored in PostgreSQL.
Building Extraction through GEOBIA and SPOR
In this section, the extraction of buildings from remotely
sensed imagery is presented. The aim was to develop a prob-
lem solving strategy to represent the implicit relationships
Figure 1. Expressions supported by
spor
.
Figure 2.
spor
reasoning process (computeMembershipValue). The process is called recursively each time the membership value of an
object to a class needs to be determined.
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
June 2015
493
