ONTOLOGICAL MODEL FOR CHARACTER RECOGNITION BASED ON SPATIAL RELATIONS

Signal & Image Processing : An International Journal (SIPIJ) Vol.4, No.3, June 2013
DOI : 10.5121/sipij.2013.4310 113
ONTOLOGICAL MODEL FOR CHARACTER
RECOGNITION BASED ON SPATIAL RELATIONS
Aicha Eutamene, Mohamed Khireddine Kholladi and Hacene Belhadef
MISC Laboratory, NTIC Faculty, University of Constantine 2, Algeria
{aicha.eutamene, kholladi, hacene_belhadef}@umc.edu.dz
{aicha.eutamene, kholladi, hacene_belhadef}@yahoo.fr
ABSTRACT
In this paper, we present a set of spatial relations between concepts describing an ontological model for a
new process of character recognition. Our main idea is based on the construction of the domain ontology
modelling the Latin script. This ontology is composed by a set of concepts and a set of relations. The
concepts represent the graphemes extracted by segmenting the manipulated document and the relations are
of two types, is-a relations and spatial relations. In this paper we are interested by description of second
type of relations and their implementation by java code.
KEYWORDS
Ontology, Grapheme, Character Recognition, Semantic Gap & Spatial Relations
1. INTRODUCTION
Handwritten character recognition is still a challenging problem for many languages like Farsi,
Chinese, English, etc. Developing robust optical character recognition techniques would be very
rewarding in today technology. Huge amount of research in this area has also contributed to solve
other open problems in pattern recognition [1], such as the character recognition in old documents
and word spotting technology.
The general goal of document recognition, whether printed or handwritten, is to turn it into a
representation understandable and exploitable by machine. The process of recognition is not
always easy as long as the content of the documents can have multiple representations.
Character recognition remains one of the vital research areas mainly over the past three decades.
In recent years, ontologies have proven to be an important and useful tool for the representation,
sharing and reuse of knowledge. Obviously also through ontology languages that can express a
rich semantic and provide reasoning capabilities.
Knowledge representation under ontological form provides the following advantages: (1) it offers
the possibility to establish a common understanding of the field of knowledge considered, (2) it
allows representing knowledge in a form convenient for automated processing information and
analysis systems, and provides an opportunity for the acquisition and accumulation of new
knowledge and multiple use of knowledge. The association between graphic data and ontologies
then allows to software- agents dash profit of the knowledge represented in ontologies to better
exploit the images. So whatever the wealth of content from documents, this last remains
insufficient to help the process to accomplish its task, so, an annotation step of the document is
required to add additional information leading to the proper functioning of this process. The
annotation of an image through the construction of ontologies is the main tool for associating

114
semantics to an image and allows the use of research methods more powerful and capable of
responding to complex queries.
The current revolution in the World Wide Web by using the notion of ontology has influenced in
many domains of scientific research such that the domains that try bring a semantic to data and
treatment; character recognition is one of these domains. In this paper, we tried to devise a new
approach to character recognition (handwritten or printed) assisted by a domain ontology. The
main idea is to model the domain of character recognition by a domain ontology, whose concepts
are graphemes (defined hereinafter) extracted from a segmentation step and extraction primitive,
these graphemes are interconnected by their spatial relationships showing their inter-location in
the document and having both intrinsic and extrinsic properties describing their forms.
The use of ontology in such a process is justified by all the benefits of the latter, in order to
improve the quality of results by introducing the concept of semantic and inference on facts that
already exists. Our primary objective by given this proposition is to reduce and bridge the
semantic gap between low-level knowledge provided by the image in the form of pixel and the
high-level knowledge extracted and enriching the image. In the domain that we studied: the
lower-level consists of several segments or primitives, characterized by low-level descriptors. The
top level includes a generic model of knowledge "ontology" and an instantiated model of
knowledge, valid for the vocabulary of the language written in the image being processed. So our
approach is based on an ontological modelling of all graphemes can constitute the Latin script and
spatial relationships that may exist between these graphemes to build letters of vocabulary, our
studies focused primarily on the Latin printed capital-letter alphabet to test the feasibility of our
approach, but it can be generalized to other forms of writing in adopting new forms of graphemes
and creating new spatial relations, specific of features vocabularies may exist between concepts.
The reliability of our approach depends on a crucial step is to segment the text document to a set
of graphemes, so a good segmentation is already an important step for good recognition, because
it is always problematic segmentation and that is why we preferred to manual segmentation by an
expert guided to maximize the percentage of expected results. After segmentation and feature
extraction, we proposed a normalization step of morphological primitives that based on a
classification (each class represents a type of graphemes), this classification will allow us to unify
and to appoint all primitives similar under the name of single grapheme, this last going to
facilitate us the instantiation of ontology concepts (a concept represents a class of primitives).
2. ONTOLOGY
The term ontology originates from philosophy. In that context, it is used as the name of a subfield
of philosophy, namely, the study of the nature of existence (the literal translation of the Greek
word ΟντολογιαΟντολογιαΟντολογιαΟντολογια), the branch of metaphysics concerned with identifying, in the most general
terms, the kinds of things that actually exist, and how to describe them. For example, the
observation that the world is made up of specific objects that can be grouped into abstract classes
based on shared properties is a typical ontological commitment [2].
However, in more recent years, ontology has become one of the many words hijacked by
computer science and given a specific technical meaning that is rather different from the original
one. Instead of “ontology” we now speak of “an ontology”. For our purposes, we will uses
T.R.Gruber’s [3] definition, later refined by R.Studer, An ontology is an explicit and formal
specification of a conceptualization.
In general, an ontology describes formally a domain of discourse. Typically, an ontology consists
of a finite list of terms and the relationships between these terms. The terms denote important

115
concepts (classes et objects)of the domain. For example, in a university setting, staff members,
students courses, lecture theaters, and disciplines are some important concepts.
The relationships typically include hierarchies of classes. A hierarchy specifies a class C to be a
subclass of another class C if every object in C is also included in C.
Apart from subclass relationships, ontologies may include information such as:
• Properties
• Value restrictions
• Disjointness statements
• Specification of logical relationships between objects
In the context of the Web, ontologies provide a shared understanding of a do main. Such a shared
understanding is necessary to overcome differences in terminology. One application’s zip code
may be the same as another application’s area code. Another problem is that two applications
may use the same term with different meanings.
Ontologies are useful for the organization and navigation of Web sites. Many Web sites today
expose on the left-hand side of the page the top levels of a concept hierarchy of terms. The user
may click on one of them to expand the subcategories.
Also, ontologies are useful for improving the accuracy of Web searches. The search engines can
look for pages that refer to a precise concept in an ontology instead of collecting all pages in
which certain, generally ambiguous, keywords occur. In this way, differences in terminology
between Web pages and the queries can be overcome.
In addition, Web searches can exploit generalization/specialization information. If a query fails to
find any relevant documents, the search engine may suggest to the user a more general query. It is
even conceivable for the engine to run such queries proactively to reduce the reaction time in case
the User adopts a suggestion. Or if too many answers are retrieved, the search engine may suggest
to the user some specializations.
In Artificial Intelligence (AI) there is a long tradition of developing and using ontology
languages. It is a foundation Semantic Web research can build upon. At present, the most
important ontology languages for the Web are the following:
• XML provides a surface syntax for structured documents but imposes no semantic
constraints on the meaning of these documents.
• XML Schema is a language for restricting the structure of XML documents.
• RDF is a data model for objects (“resources”) and relations between them; it provides a
simple semantics for this data model; and these data models can be represented in XML
syntax.
• RDF Schema is a vocabulary description language for describing properties and classes
of RDF resources, with a semantics for generalization hierarchies of such properties and
classes.
• OWL is a richer vocabulary description language for describing properties and classes,
such as relations between classes (e.g., disjointness), cardinality (e.g. “exactly one”),
equality, richer typing of properties, characteristics of properties (e.g., symmetry), and
enumerated classes [2].

116
3. SENSORY AND SEMANTIC GAP
Visual similarity does not necessarily a semantic similarity, Example: two regions in an image of
the same color does not mean they represent the same region or the same object. This difference
between the conceptual level between machine, which knows only the pixelique data (bridge,
line, curve, etc..) And the user who can interpret them (house, bridge, etc.) is called semantic gap.
It is the recognition system to bridge this gap by proposing a high-level interpretation of low-level
data.
The sensory gap is defined as "the gap between the objects in the real world and the information
contained in a description (computer) derived from recording the scene." It is the projection of a
reality, 3D and often continues in a 2D discrete computer representation. This gap is to be
accepted by researchers working on 2D images, or repelled by researchers working on
stereoscopic or 3D images (see figure 1).
The semantic gap is the most difficult to treat. For several years, researchers have revolved
around this gap without actually naming it, what is done today. The semantic gap is defined as
"the lack of concordance between the information that can be extracted from visual data and the
interpretation of these data for a user in a given situation." This gap is more or less the same
problem as linking lowlevel treatments and highlevel treatments, except that now it is clearly seen
as a problem of information management and not only as a control problem[4].
In recent years, the gap between semantic concepts and low-level numerical features retained
much attention from the scientific community, and remains one of the major challenges in the
field of computer vision. To reduce this semantic gap, we have adopted the proposed approach in
[6] [7] described by a character recognition system based on ontology and a segmentation of the
document processed into graphemes. The types of graphemes that we have adopted are classed in
four classes according the classification of philipe Coueignoux [8] (see figure 5).
The ontology plays a role of a bridge between the information represented in the image and
knowledge of the expert. Figure 2 illustrates how the information of the image might take a form
based on ontologies (left), as well as the knowledge of the expert (right).
Figure 1. The sensorial gap between the observed world and the acquired image. The semantic
gap between computer analysis and human interpretation of the image [5].

117
Figure 2. Ontological bridge between image and knowledge
5. MODELING OF SPATIAL RELATIONS
Various approaches underlying the modelling of spatial relations, which is a heterogeneous and
interdisciplinary field. To model the reality, it is not sufficient to define geographic entities. It
should also define spatial relationships between these entities. These spatial relationships are
important, especially for knowledge representation, spatial reasoning, and of course, the spatial
integrity constraints. They can be classified as follows [9] :
- Relationships based on a spatial order linked to a definition of origin or axis or (eg, direction) ,
- The topological relations that describe neighbourhoods and set-properties (eg, disjoint, adjacent
intersects)
- Metric relationships related to distance.
Figure 3, shows the tree of taxonomy of these relationships and table 3 illustrates the detail of
RDF/OWL code for each class.
Figure 3. relationships taxonomy
Image information in
pixel matrix
Image information in
ontology
Expert knowledge as
ontological form
Expert knowledge
IMAGE KNOWLEDGE
Topological
relation
Spatial-relation
Metric
relation
Included
relation
Adjacent
relation
Direction
relation
Distance
relation
Binary direction
relation
Ternary direction
relation
atRight atLeft below above
Represents a subclass
Represents a class

118
A spatial relationship allows describing the relative positions between two symbols. For example,
given the expression ab, the symbols a and b are connected by the relation left/right. While they
are connected by the relation (top-right/bottom-left) in the expression ab
. In most systems
proposed a relationship bears more structural sense a semantic meaning (logical relation). For
example, the structural relationship in ab
means mathematically that a to the power b [10].
Figure 4 shows an extract of the spatial taxonomy implemented under the Protégé2000 editor.
This tree shows the spatial relationships of the Latin alphabet, but we can enrich it and reuse it,
for other types of alphabet (example: Arabic, Chinese, etc).
Figure 4. Relationships taxonomy under Protégé2000
The types of relationships that we addressed in our study are the following relations:
• adjacency between two segments or two objects of interest or two semantic objects
(distinguishing adjacencies top, bottom, right and left): "a Bar is adjacent to a Trunk and is
located to the right of it (the case of letter 'L'). "
• Neighbourhood between two segments or two objects of interest or two semantic objects
without contact between them (distinguishing cases north, south, east, west) "a Point is located
north of Trunk" (the case of the letter 'i') (see figure 7).
Figure 5. Classification of graphemes by philipe Coueignoux
According to the classification of philipe Coueignoux, we created taxonomy of graphemes, where
each of them is represented by a concept bearing his name. Figure 5 shows an excerpt of this

119
taxonomy published under the ontologies Editor: Protégé2000 (The Protégé Ontology Editor and
Knowledge Acquisition System: http://protege.stanford.edu/)
Figure 6. Taxonomy of graphemes
Figure 7, shows an example of segmented character in two graphemes such as "Small_Trunk" and
"Point" who they are connected by a spatial relation : “Above”, you can also note that this last can
have an inverse relationships which is “Below”, it is possible to implement this type of relations
with Protégé2000.
i
Figure 7. graphemes and spatial relations composing the character ”i”
and the RDF/OWL code of this relationship is mentioned here by this code:
<owl:ObjectProperty rdf:ID="Below">
<owl:inverseOf rdf:resource="#Above"/>
</owl:ObjectProperty>
PointSmall_Trunk
Above
Below

120
Figure 8, shows via the plugin Ontoviz of the protégé2000 editor, an example of an instantiation
of the concepts "Small_Trunk" and "Point" and the relationships "above, Below" for building of
character "i".
Figure 8. An example of ontology instantiation (Case of character : i)
6. IMPLEMENTATION AND RESULT
Protégé has an open architecture that allows programmers to integrate plug-ins, which can appear
as separate tabs, specific user interface components (widgets), or perform any other task on the
current model. The complete source code for most of the examples can be found in the
edu.stanford.smi.protegex.owlx.examples package in the Protégé-OWL source code [11]. if you
use Eclipse or Netbeans as a Java IDE, it must select the Protégé installation folder as your
project home, then add all the JAR files from the installation to your project class path.
Protégé-OWL API is centered around a collection of Java interfaces from the model package that
provide access to the OWL model and its elements (classes, properties, and individuals). The
most important interface is OWLModel which provides access to the top-level container of the
resources in the ontology so you can use OWLModel to create, query, and delete resources of
various types and then use the objects returned by the OWLModel to do specific operations [12].
Table 1, shows the example for creating Spatial-relation class with java code.
Table 1. Java code for creating Spatial-relation class:
import edu.stanford.smi.protegex.owl.model.OWLModel;
import edu.stanford.smi.protegex.owl.model.OWLNamedClass;
import edu.stanford.smi.protegex.owl.ProtegeOWL;
public class BuildSpatialTaxonomy{
i

121
public static void main(String[] args) {
OWLModel owlModel= ProtegeOWL.createJenaOWLModel();
owlModel.getNamespaceManager().setDefaultNamespace("http://SpatialTaxonomy.com#");
OWLNamedClass spatialClass = owlModel.createOWLNamedClass
("Spatial-relation");
System.out.println("Class URI: " + spatialClass.getURI());
}
}
In the table 2, we have outlined the most important primitives that allow read and browse the
OWL ontologies. The package "com.hp.hpl.jena.ontology" provides a set of classes and methods
to manipulate and access to this type of ontology.
Table 2. Primitives of manipulating the OWL ontologies
Role Statement
Creating an ontology model m.
OntModel m=ModelFactory.createOntologyModel
(OntModelSpec.OWL_MEM,null);
Load and creation a copy of
ontology
m.getDocument Manager. addAltEntry()
Reading of ontology m.read ( );
Browse the ontology classes.
Give all classes of ontology
for (Iterator i= m.listClasses ( ) ; i.hasNext());
{OntoClass cls= (OntClass) i.next();
Local name of class
getLocalName ()
System.out.println(cls.getLocalName);
Browse the super classes of cls
for (Iterator i =cls. listSuper.Classes( ) ; i .haslNext())
{ Sysem.out.print(((OntClass)i.next()).getLocalName()
+“ ” ); }
Browse the subclasses of cls
for (Iterator i =cls. listSub. Classes( ) ; i .haslNext())
{ Sysem.out.print(((OntClass)i.next()).getLocalName()
+“ ” );}
Table 3. Implementation code RDF/OWL of relationships taxonomy
Concepts of relationship
Taxonomy
Code Rdf/OWL
Spatial-relation
<owl:Class rdf:ID="Spatial_Relation"/>
</owl:Class>
<owl:Class rdf:ID="Metric_Relation">

122
Metric relation <rdfs:subClassOf
rdf:resource="#Spatial_Relation"/>
</owl:Class>
Direction relation
<owl:Class rdf:ID="Direction_Relation">
<rdfs:subClassOf
rdf:resource="#Metric_Relation"/>
</owl:Class>
Distance relation
<owl:Class rdf:ID="Distance_Relation">
<rdfs:subClassOf
rdf:resource="#Metric_Relation"/>
</owl:Class>
Binary relation
<owl:Class rdf:ID="Binary_Relation">
<rdfs:subClassOf
rdf:resource="#Direction_Relation"/>
</owl:Class>
atRight
<owl:Class rdf:ID="atRight">
<rdfs:subClassOf
rdf:resource="#Binary_Relation"/>
</owl:Class>
atLeft
<owl:Class rdf:ID="atLeft">
<rdfs:subClassOf
</owl:Class>
below
<owl:Class rdf:ID="below">
<rdfs:subClassOf
</owl:Class>
above
<owl:Class rdf:ID="above">
<rdfs:subClassOf
</owl:Class>
Adjacent relation
<owl:Class rdf:ID="Adjacent_Relation">
<rdfs:subClassOf
rdf:resource="#Topological_Relation"/>
</owl:Class>
Included relation
<owl:Class rdf:ID="Inclued_Relation">
<rdfs:subClassOf
rdf:resource="#Topological_Relation"/>
</owl:Class>
Topological Relation
<owl:Class rdf:ID="Topological_Relation">
<rdfs:subClassOf
rdf:resource="#Spatial_Relation"/>
</owl:Class>
7. CONCLUSIONS
In this paper, we presented a new paradigm for character recognition problem based on the
domain ontology that is characterised by description of spatial relations between graphemes.
These last are represented by ontological concepts and the relations linking them are of spatial
type describing their spatiality in the document processed. In this article we haven’t published all
type of relations but only the most important.

123
The development of Ontologies in this area can be used to provide image analysis automation
support and efficient use of modern methods and techniques for image analysis and pattern
recognition.
In the future work, we plan to expand and enrich the taxonomy of relationships with other types
of relationships and add SWRL inference rules to strengthen the reasoning on our ontology. We
plan to use the matching operations with external resources such as WordNet by using dedicated
similarity measures in the step of post-treatment.
REFERENCES
[1] Ali Borji • Mandana Hamidi • Fariborz Mahmoudi, (2008) “Robust Handwritten Character
Recognition with Features Inspired by Visual Ventral Stream”, Neural Processing Letters 28(2): 97-
111.
[2] Grigoris Antoniou and Frank van Harmelen, (2004) “A Semantic Web Primer”, TLFeBOOK, The
MIT Press.
[3] T.R. Gruber, (1993) “A Translation approach to Portable Ontology Specification Knowledge
Acquisition”, International Journal of Knowledge Acquisition for Knowledge-based systems, Vol. 5,
Num. 2.
[4] A.W.M. Smeulders, M. Worring, S.Santini, A.Gupta and R.Jain, (2000) “Content-Based Image
Retrieval at the End of the Early Years”. IEEE Trans. on Patt. Anal. and Machine Intell., Vol. 22,
Num. 12, pp. 1349-1380.
[5] Alain Boucher and Thi-Lan Le, (2005) “Comment extraire la sémantique d’une image ? ”, SETIT
2005, 3rd International Conference: Sciences of Electronic, Technologies of Information and
Telecommunications.
[6] Hacene Belhadef and Aicha Eutamene, (2012) “Ontology and character recognition : New axis of
research, International journal of academic research Vol. 4. Num. 2.
[7] Aicha Eutamene, Hacene Belhadef and M. K. Kholladi, (2011) “New Process Ontology-Based
Character Recognition”, LNCS Series of Communications in Computer and Information Science, Vol.
240.
[8] Ph. Coueignoux, (1983) “Approche structurelle de la lettre In: Langue française”, N°59, pp. 45-67.
[9] Marc Souris, “Les principes et algorithmes des systèmes d'information géographique
L'exemple du système SavGIS”, Chapter 7 : “Contraintes d"'intégrité spatiale et saisie graphique”,
[10] Ahmad Montaser AWAL, (2010) “Reconnaissance de structures bidimensionnelles : Application aux
expressions mathématiques manuscrites en-ligne”, Thèse de Doctorat de l’Ecole polytechnique de
l’Université de Nantes.
[11] Protege-OWL API Programmer's Guide, URL:
http://protegewiki.stanford.edu/wiki/ProtegeOWL_API_Programmers_Guide
[12] Working with OWL Models, URL: http://protegewiki.stanford.edu/wiki/ProtegeOWL_API_Basics

124
Authors
Aicha Eutamene is Phd student at NTIC Faculty, University of Constantine 2-Algeria.
Mohamed Khireddine Kholladi, Professor at NTIC Faculty, University of Constantine 2-Algeria, President
of MISC Laboratory
Hacene Belhadef, Associate professor at NTIC Faculty, University of Constantine 2-Algeria, Member of
MISC Laboratory (corresponding author email: hacene_belhadef@yahoo.fr)

ONTOLOGICAL MODEL FOR CHARACTER RECOGNITION BASED ON SPATIAL RELATIONS

More Related Content

What's hot (18)

Viewers also liked (19)

Similar to ONTOLOGICAL MODEL FOR CHARACTER RECOGNITION BASED ON SPATIAL RELATIONS (20)

Recently uploaded (20)

ONTOLOGICAL MODEL FOR CHARACTER RECOGNITION BASED ON SPATIAL RELATIONS