PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
June 2015
491
A Fuzzy Spatial Reasoner for
Multi-Scale GEOBIA Ontologies
Argyros Argyridis and Demetre P. Argialas
Abstract
In Geographic Object-Based Image Analysis (
GEOBIA
) an image is
partitioned into objects by a segmentation algorithm. These objects
are then classified into semantic categories based on unsupervised/
supervised methods, or knowledge-based methods, such as an
ontology. The aim of this paper was to develop a SPatial Ontology
Reasoner (
SPOR
) to allow the development of
GEOBIA
ontologies
by employing fuzzy, spatial, and multi-scale representations,
with time efficiency. An enhanced version of the Web Ontology
Language 2 (
OWL
2) with fuzzy representations was adopted and
expanded to represent fuzzy spatial relationships within the frame-
work of
GEOBIA
. Segmentation results are stored within PostgreSQL.
An ontology described the class/subclass hierarchy and class defi-
nitions.
SPOR
integrated PostgreSQL and the ontology, to classify
the objects. To demonstrate the framework, a QuickBird image was
employed for building extraction. Accuracy assessment indicated
that 87 percent of building rooftops were detected.
Introduction
In GEographic Object-Based Image Analysis (
GEOBIA
) an
image is partitioned into primitives (segments), which then
are classified into semantic categories by employing Standard
Nearest Neighbor, fuzzy inferencing, or advanced machine
learning techniques (Baatz and Sch pe, 2000; Benz
et al.
,
2004; Blaschke
et al.
, 2008; Hay and Castilla, 2008; Tzotsos
and Argialas, 2008; Blaschke and Strobl, 2010; Mallinis
et
al.
, 2013; Blaschke
et al.
, 2014). However, to extract complex
landscape components in terms of spectral, geometric, and
rich spatial relationships, it is required the representation of
expert knowledge into a problem-solving strategy through
an establish-and-refine-paradigm within the environment of
a knowledge representation system (Argialas and Harlow,
1990; Argialas
et al.
, 2013). This requires the application of
heuristic rules derived from knowledge stored in books, pho-
to-interpretation manuals, relative work in the field, and per-
sonal experience of the phenomenon (Argialas and Harlow,
1990; Arvor
et al.
, 2013). To take advantage of this symbolic
knowledge within an automated image analysis system, it is
required to be formalized into a computer-conceivable form.
Thus, a semantic gap arises between the high-level semantics
employed by the experts to describe the phenomenon (Vege-
tation has high infrared reflectance values) and the numerical
low-level information extracted from data (
NDVI
values greater
than 0.25). To address this problem, methods are required to
identify optimal features to discriminate between evaluated
classes and to explicitly specify the knowledge of the experts
on the evaluated classes (Belgiu
et al.
, 2014). To this end,
rule-based systems and ontologies offer potential for knowl-
edge formalization (Lüscher
et al.
, 2009; Belgiu
et al.
, 2014).
In recent years, ontologies have become popular as a
means of representing machine-readable knowledge. An on-
tology is defined as a formal, explicit specification of a shared
conceptualization (Gruber, 1995). Ontologies allow for cap-
turing the semantics of the domain concepts into knowledge
organization systems that can be easily reused and extended
(Belgiu and Lampoltshammer, 2013). In Guarino (1997),
ontologies were classified based on their detail as top-lev-
el (describing generic concepts), domain (describing the
knowledge of a certain field), task (describing generic tasks),
and application ontologies (describing concepts related to a
certain field and related tasks). Domain, task, and application
ontologies need to be aligned with the top-level ontology to
ensure collaboration with other domain applications.
Ontologies can help solve the problem of the semantic gap
towards the implementation of an automatic image recogni-
tion system based on
GEOBIA
that is able to bridge the sym-
bolic information derived from the experts and the numerical
information extracted from the image (Blaschke
et al
., 2014).
Ontology-based recognition consists of classifying an image
object as an instance of a specific type if it satisfies all of the
constraints defined in the ontology for that object type (Arvor
et al
., 2013). The development of ontology-based recognition
requires (a) formalizing the symbolic knowledge of an expert,
of a specific image object type in an ontology, and (b) associ-
ating this knowledge with image segments described through
annotations that are based on the same ontology.
Ontology studies have already been conducted in
GIS
,
scene analysis, remote sensing, and
GEOBIA
. In
GIS
, ontologies
have been successfully applied to perform knowledge repre-
sentation. Torres
et al
. (2005) employed ontologies to describe
the semantic content of topographic and thematic maps.
Lutz and Klien (2006) used an ontology to explicitly specify
and formalize the meaning of the domain concepts into a
machine-readable language that enabled spatial information
retrieval on a semantic level. Zhan
et al
. (2008) developed a
framework to retrieve spatial information, based on spatial
relation and geometric relation ontologies. Lüscher
et al
.
(2009) developed an ontology-driven approach for cartograph-
ic pattern recognition in support of map generalization.
In the field of scene analysis, ontologies have been also em-
ployed to extract scene content. Wang
et al
. (2006) established a
method for retrieving scene imagery, by developing ontologies
which combined text annotation and image features. Hudelot
et al
. (2008) organized spatial relationships in an ontology
representing topological relations (adjacency, inclusion, etc.)
and metric relations which further contained distance relations
(far/close distance, etc.), and directional relations (right to,
left to, etc.). This ontology was linked with an upper ontology,
describing brain entities, and spatial reasoning was performed
to recognize various elements of the brain. In Bannour and
Hudelot (2014) a multi-stage reasoning approach was de-
veloped, to perform semantic annotation/tagging on a target
School of Rural and Survey Engineering, Remote Sensing
Laboratory, National Technical University of Athens, Iroon
Polytechneiou 9 Zografou/Athens, Greece, Postal Code 15780
(
).
Photogrammetric Engineering & Remote Sensing
Vol. 81, No. 6, June 2015, pp. 491–498.
0099-1112/15/491–498
© 2015 American Society for Photogrammetry
and Remote Sensing
doi: 10.14358/PERS.81.6.491
