Conference PaperPDF Available

Hybrid Ontology and Visual-Based Retrieval for Cultural Heritage Multimedia Libraries

Authors:

Abstract and Figures

Nowadays, an increasingly growing demand for the creation of digital multimedia libraries is arising, as huge amounts of digital visual content are becoming available. The content that resides in these libraries should be easily retrievable and classified in order to be fully accessible. This paper introduces a hybrid multimedia retrieval model accompanied by the presentation of a search engine that is capable of retrieving visual content cultural heritage multimedia libraries as in three modes: (i) based on semantic annotation with the help of an ontology; (ii) based on the visual features with a view to finding similar content; and (iii) based on the combination of these two strategies. The main novelty is the way in which these two co-operate transparently during the evaluation of a single query in a hybrid fashion, making recommendations to the user and retrieving content that is both visually and semantically similar.
Content may be subject to copyright.
Hybrid Ontology and Visual-based Retrieval for Cultural Heritage Multimedia
Libraries
Stefanos Vrochidis, Charalambos Doulaverakis, Lambros Makris, Anastasios Gounaris,
Evangelia Nidelkou, Ioannis Kompatsiaris and Michael G. Strintzis
Informatics and Telematics Institute
Thessaloniki, Greece
{stefanos, doulaver, lmak, gounaris, nidelkou, ikom, strintzi}@iti.gr
Abstract
Nowadays, an increasingly growing demand for the cre-
ation of digital multimedia libraries is arising, as huge
amounts of digital visual content are becoming available.
The content that resides in these libraries should be eas-
ily retrievable and classified in order to be fully accessible.
The contribution of this paper is the introduction of a hy-
brid multimedia retrieval model accompanied by the pre-
sentation of a search engine that is capable of retrieving
visual content cultural heritage multimedia libraries as in
three modes: (i) based on their semantic annotation with
the help of an ontology; (ii) based on the visual features
with a view to finding similar content; and (iii) based on the
combination of these two strategies. To achieve this, the re-
trieval model is composed of two different parts, a low-level
visual feature analysis and retrieval and a high-level ontol-
ogy infrastructure. The main novelty is the way in which
these two co-operate transparently during the evaluation of
a single query in a hybrid fashion, making recommenda-
tions to the user and retrieving content that is both visually
and semantically similar.
1. Introduction
Digital multimedia Libraries are organized in order to
store and manage huge amounts of audiovisual content for
universal remote access.
The huge amount of visual digitized information has to
be structured and annotated in the proper way in order to
be accessible and retrievable. To provide functionalities
for the manipulation and knowledge retrieval from such vi-
sual content, a key aspect is the development of more effi-
cient search engines to handle image and video files. To
date, two main approaches to image search engine tech-
niques have been proposed, namely annotation-based and
content-based. Several variants of annotation-based multi-
media search engines have been proposed. Some of them
assume manual annotation reducing visual search to infor-
mation retrieval [1], while others provide support for auto-
matic annotation.
This search approach has benefitted significantly from
the advances in the Semantic Web and ontologies (e.g., [2]),
so that annotations can have well-defined semantics. On-
tologies are “an explicit specification of a conceptualiza-
tion” [3], and they guarantee firstly a shared understanding
of a particular domain, and secondly, a formal model that is
amenable to unsupervised, machine processing.
However, annotation and semantic-based search are of-
ten insufficient when dealing with visual content. To tackle
this problem, a second complementary approach has been
devised: content-based search. The core idea is to apply
image processing and feature extraction algorithms to the
visual content and extract low-level visual features, such as
color layout and edge histogram [4].
This paper focuses on a hybrid retrieval model by com-
bining in an novel way the content and annotation based
approaches. A search engine has been developed imple-
menting this model using images from the culture domain.
Regarding content-based search, the engine employs state-
of-the-art techniques, which involve automatic segmenta-
tion of 2D visual content and MPEG-7 features extraction,
while the novel, hybrid search functionality is capable of
extending either the content-based search by making user
suggestions of additional, potentially interesting results.
The remainder of the paper is structured as follows. Sec-
tion 2 introduces the retrieval model while the evaluation
procedure including the presentation of the search engine is
presented in Section 3. Examples and results of the hybrid
engine appear in Section 4.
Eventually, section 5 concludes the paper.
2 Hybrid Retrieval of Visual Content
As mentioned previously, the search engine described
hereby supports three modes of queries and retrieval of im-
ages and video, namely
1. content-based retrieval,
2. ontology-based retrieval, and
3. hybrid retrieval, which builds upon the combination of
the two aforementioned methods.
2.1 Combining visual and semantic infor-
mation
The main objective behind our retrieval model is to al-
lowa user to complementa query primarily addressed using
one of the visual or semantic mode with the other. Starting
with one mode, information arising from the complemen-
tary mode is used to enhance the results. The additional
results presented are considered to be a set of recommen-
dations for the user by broadening the desirable query and
they are generated in a transparent way.
The mathematical model of the hybrid retrieval system
is described below for both cases. Lets assume that the
function Sem(data
rdf
, q
sem
) is producing the desired out-
put given the data and the query based on the semantic data
formed in RDF
1
language by retrieving the results from the
Knowledge Base:
Res
sem
= Sem(data
rdf
, q
sem
) (1)
where data
rdf
are the metadata stored in RDF in the
Knowledge Base and q
sem
is the query string in RDFQL
or SeRQL.
In a similar way, the function: V is(data
desc
, q
vis
) out-
puts the results from content-based search using as data
the extracted descriptors of the multimedia content and the
proper input from the user.
Res
vis
= V is(data
desc
, q
vis
) (2)
where data
desc
represent the extracted descriptors of the
multimedia content and q
v is
represents the desirable input
(i.e one or a set of images) for which visually matching
content expected to be retrieved and displayed. The func-
tion V is(data
desc
, q
vis
) outputs Res
v is
in a specific rank-
ing based on the similarity coefficient which derives from
the calculation of the distances of the extracted descriptors
for the objects included in the query.
1
http://www.w3.org/RDF/
Subsequently, two cases of hybrid search are defined:
(i) the visual search, where the system, given the desir-
able query, produces visually similar results with the ini-
tial object accompanied by a set of recommendations de-
riving from the transparent semantic query that the visual
results produce; and (ii) the semantic search, where a user
can submit a query by browsing the ontology fields and ac-
quire the results that illustrate the content which satisfies
the constraints of the query complemented by recommen-
dations based on visual similarity of the initial results.
In the case of visual search the output consists of two sets
of results: the initial results produced by Res
v is
and the set
of recommendations Rec
sem
based on semantic search and
given by:
Rec
sem
= Sem(data
rdf
, ResT oQ
sem
(Res
v is
)) <=>
(3)
Rec
sem
= Sem(data
rdf
, ResT oQ
sem
(V is(data
desc
, q
vis
)))
(4)
where the function ResT oQ
sem
creates a new query based
on the first set of the results in order to retrieve the semanti-
cally related content. The aforementioned function exploits
the initial set of results by processing the ontology fields of
every output object in order to define the semantic anno-
tation which is mostly shared by these results. The query
produced leads to a search for content that shares the spe-
cific common value in the chosen ontology fields with the
results.
The final set of results Res is the set of results from vi-
sual similarity Res
v is
enhanced by the recommendation re-
sults Rec
sem
:
Res = Res
vis
Rec
sem
. (5)
On the other hand, when a Semantic search occurs the
results that are produced consist of: the first set provided by
Res
sem
and the second set Rec
vis
illustrating the recom-
mendations:
Rec
v is
= Sem(data
desc
, ResT oQ
vis
(Res
sem
)) <=>
(6)
Rec
v is
= V is(data
desc
, ResT oQ
vis
(Sem(data
rdf
, q
sem
)))
(7)
where the function ResT oQ
v is
constructs a query taking
into account the visual features of the initial results. The
algorithm used for this function produces descriptors of an
average hypothetical object by averaging the descriptors of
the results.
The final set of results is:
Res = Res
sem
Rec
v is
. (8)
2.2 Content-based Retrieval
In this retrieval mode, described by (2), users are able to
perform a visual-based search by taking advantage of low-
2
Segmentation
Algorithm
Shot
segmentation
Descriptor
extraction
Keyframe
extraction
Descriptor
DB
2D
content
Image
Video
Figure 1. The offline image and video content
analysis process.
level multimedia content features. The retrieval system can
handle 2D still image and potentially video. In this mode,
the user provides, as the input query, an example of the mul-
timedia content she or he is interested in, and, based on the
extracted descriptors of the input and the indexed offline-
generated descriptors of the content repository, the system
performs a visual similarity-based search and the relevant
results are retrieved.
Figure 1 illustrates the 2D image analysis with potential
application to video files. More specifically, analysis of 2D
images is performed in a two-step fashion. To enable mean-
ingful region detection in the available cultural heritage im-
ages collections, a segmentation process takes place using
the approach described in [7]. The second step in analy-
sis involves low-level feature extraction from the resulting
regions of the segmentation mask and also from the whole
image itself. For this purpose, the MPEG-7 features were
selected as they represent the state of the art in low-level
visual descriptors. For the extraction, the MPEG-7 eXperi-
mentation Model (MPEG-7 XM) [8] was applied.
This procedure could be extended or video analysis. In
such a case the video stream is firstly divided into shots us-
ing the method described in [9]. For each detected shot, a
keyframe is extracted which is treated as a compact repre-
sentation of the entire shot. This keyframe is then analyzed
as in the still image case.
2.3 Ontology-based Retrieval
This search mode, described by (1), is more appropriate
for the cases in which the user knows to an adequate de-
gree of confidence the semantic annotation of the material
user provides constraints on the concepts of the five ontolo-
gies of Section 3.2. During search time, the system retrieves
the semantically connected content according to users selec-
tions. The system can handle complex queries that require
the combination of multiple concept-based search criteria
and thus can retrieve different cultural items that share com-
Collection Ontologies
Annotation Ontology
Search Criteria
Class Hierarchy of selected concepts
}
Possible values
of selected
concepts to
formulate the
search criteria
Figure 2. Search engine interface
mon data. Figure 2 illustrates the application of this tech-
nique by presenting a proper interface for ontology based
search.
3 Search Engine and Evaluation
Corpus
The evaluation procedure took place with the employ-
ment of a search engine based on the aforementioned re-
trieval model and it is capable of retrieving cultural visual
content. The design of the GUI and the ontology browser
plays a significant role for ontology-based retrieval. As
shown in Figure 2, the GUI of our search engine provides a
view of the ontologies, enabling the browsing through their
structure and hierarchy; Apart from the ontology-based re-
trieval, the search engine supports content-based queries
in order to produce results depending on visual similarity.
Combining the two techniques, the search engine is capa-
ble of providing the hybrid searching functionality as was
described in detail in Section 2.1.
3.1 Visual Content
The main content provider is the Center for Greek and
Roman Antiquity (KERA)
2
, which offers a large collection
of inscriptions and coins , accompanied with detailed doc-
umentation. Furthermore, a rich collection of Greek paint-
ings is provided by the Greek museum: Teloglion Founda-
tion of Art
3
while a large collection of photographs is of-
fered by Alinari Photographic Archives
4
.
2
http://www.eie.gr/nhrf/institutes/igra/index-en.html
3
http://web.auth.gr/teloglion/
4
http://www.alinari.com/
3
hasComments
isQuotedIn
hasCurrentLocation
tookPlaceDuring
tookPlaceDuring
hasFindingConditions
hasC
rea
tion
Cond
iti
ons
is
Made
Of
hasCode
Bibliography
CreationEvent
ItemMaterial
Location
FindingEvent
Date
Comments
Code
Item
Figure 3. A graphical representation of the
concepts in the ontology and their relations.
3.2 Ontology
Cultural heritage collections are accompanied by a rich
set of annotations. However, these annotations are often
unstructured or registered in a non-standard form, usually
proprietary, for every collection, which renders them unus-
able for inter-collection searching. To overcome this prob-
lem, appropriate ontologies for the cultural heritage domain
have been defined.
Taking into account the content originally available for
our use case scenario an ontology infrastructure has been
defined to efficiently describe and represent all knowl-
edge related to each collection. The proposed architec-
ture consists of two layers and makes use of five different
RDF(S) ontologies, namely Annotation, which is generic,
and Coins, Inscriptions, Paintings and Photographs which
are specific to the collection itemsets of our scenario.
A large set of information fields inside the image anno-
tation set is common for each item, regardless of the collec-
tion that is part of. As such, it was decided to use a separate,
higher-level ontology specifically intended for representing
this kind of concepts and relations. dimensions, etc. Conse-
quently, the role of the Annotation ontology (Figure 3) is to
conceptualize and hold all common data in a structured way,
thus forming a representation standard for every collection
to be integrated with the search engine.
The properties that are specific to a collection item cat-
egory are captured by complementary ontologies; more
specifically there is a separate ontology for each category, as
the particular details that correspond to the collection items
can vary greatly for each class.
As a further step, to support interoperability of the sys-
tem with other semantic-enabled cultural heritage systems,
the aforementioned ontologies were mapped to the CIDOC-
CRM [10] core ontology which has been proposed as an
ISO standard for cultural heritage material structuring and
representation. To enable this functionality, appropriate
mappings between the concepts of our defined ontologies
and the CRM were drawn.
3.3 Ground Truth Definition
The ground truth used in order to evaluate the results of
the experiments were different for each retrieval mode.
Regardingthe content-based experimentsas ground truth
was considered the (subjective) visual similarity of the ob-
jects . More specifically the visual features which were
taken into account to prove visual similarity were the shape
and the color while the existence of visually related regions
between the objects can also be considered as factor of vi-
sual resemblance.
The results from ontology-based queries could be easily
evaluated due to the existing annotations.
The recommendations, which are results of the hybrid
mode, are considered to be results related to the initial set
of the semantic or visually based results. The definition of
the term recommendation can be subjective however it can
be defined as any result which is related to the initial output
in terms of visual or semantic similarity.
4 Results
The content-based and ontology-based modes are com-
plementary to each other, and as such, it is meaningless to
compare them directly in terms ofmetrics like precision and
recall. In this section the advanced functionalities of the hy-
brid search engine are demonstrated through use cases and
insights into the performance of the two search flavors are
provided.
4.1 Hybrid search: use cases
The hybrid search engine is capable of detecting implicit
semantic relationships between visually dissimilar images,
and extract the relevant artefacts. To demonstrate this capa-
bility, two use cases are presented in this section, which are
summarized in Figures 4 and 5.
In the first use case (first row of Figure 4), the input query
is the painting Autoprosopgrafia” (“Self-portrait”). Dur-
ing the content-based search visually similar inscriptions
are extracted (Figure 4a). As shown in the figure, the results
have severalvisualfeatures in common, such as depiction of
human faces and other portraits. To fire an ontology-based
search, the system retrieves the most common semantic fea-
ture, which happens to be the name of the painter. As such,
the system is capable of automatically returning the same
results as if the user was capable of submitting a query of
the type (in natural langauge) “Find all the artefacts that
look like the painting “Autoprosopgrafia” or all the Paint-
ings created by Pierrakos Alkis” (Figure 4b). The first set of
results shows visually similar images by illustrating mostly
paintings with human faces and portraits while the recom-
mendations include works of art of the same painter who
4
Figure 4. The first use case: (a) initial set of
results derived from visual similarity search,
(b)set of recommendations based on a com-
plementary semantic query
has created paintings of similar themes (portraits with dif-
ferent technique etc.) complementing in this way the ini-
tial results by providing paintings of similar subjects which
could not be retrieved by visual similarity due to the differ-
ent paint technique(Figure 4).
The second use case follows the opposite approach
Figure 5. The second use case: (a) initial set
of results based on a semantic query, (b) set
of recommendations includes visually simi-
lar images
where the initial query is based on ontology fields while
the set of recommendations derives from visual similarity.
In this scenario, the user searches for inscriptions dated in
the 4th century BC. The first set of results includes the in-
scriptions which belongs to the aforementioned period (Fig-
ure 5a) while the recommendations provide results visu-
ally similar with the inital set of inscriptions. As it can
be observed in Figure 5b the recommendation images in-
clude more inscriptions with similar shapes and figures with
the first set of results. Consequently the recommendations
broaden the initial query as they reveal visually similar im-
ages of potential interest that the user may not have been
aware of when submitted the query.
5
Figure 6. Precision-Recall diagram for the
content-based method.
4.2 Performance Insights
The experiments were conducted on a PC, with a P5
3.0GHz Intel CPU and 1GB RAM. The knowledge base
containing the ontological metadata is Sesame 1.2 running
a MySQL DBMS at the back-end. The dataset is consisted
of roughly 4000 images along with a complete set of se-
mantic annotations. The visual descriptors are stored in a
collocated MPEG-7 XM server.
Figure 6 shows the Precision-Recall diagram for the
content-based retrieval. The curves correspond to the mean
precision value that was measured after several retrieval
tasks. For the ontology-based search since it is based on
selecting available concepts describing the content, the esti-
mation of Precision-Recall diagram is not relevant. The av-
erage response time for the ontology-based search is 0.163
sec, while for the content-based search is 0.773 sec.
The behavior of the hybrid search is expected to com-
bine the benefits of the other two approaches providing rec-
ommendations to the user in order to broaden the query.
Precision-Recall graphs for the recommendations are not
presented as these strongly depend on the nature of the re-
trieval task, and on the objective and purpose of the user
when submitting a query.
5 Conclusions
In this paper, a novel retrieval model for handling vi-
sual and multimedia digital libraries is presented in an effi-
cient and effective manner. The search engine based on that
model adopts three methods for retrieval: two autonomous
and one combinational. The ontology-based method makes
use of the formal, logic-based representation of semantic
mark-up metadata accompanying each collection, while an
illustrative user interface is used for graphical query formu-
lation. Although the search engine dealed with 2D images
of cultural heritage content, there is the potential of exten-
sion based on the proposed model to include video content.
A notable feature of this work is its modular and exten-
sible ontology infrastructure, which provides mappings to
CIDOC-CRM in order to gain interoperability with other
ontologies from the cultural domain. The hybrid method,
which is the main contribution of this work, makes a com-
bined use of the previous two methods. Thus is capable of
offering, as a recommendation, a more complete result set
to the user, which comprises both visually and semantically
similar items, while the input query remains either solely
ontology-based or content-based.
References
[1] E. Rillof and L. Hollaar. Text Databases and Information Re-
trieval. In ACM Computing Surveys, vol. 28, no.1, pp. 133-135,
1996.
[2] P. Sinclair, S. Goodall, P. Lewis, K. Martinez, and M. Ad-
dis. Concept browsing for multimedia retrieval in the SCULP-
TEUR project. In Multimedia and the Semantic Web, held as
part of the 2nd European Semantic Web Conference, Herak-
lion, Crete, Greece, May 2005.
[3] T. Gruber. A Translation Approach to Portable Ontology
Specifications. Knowledge Acquisition, 5:199–220, 1993.
[4] A. Smeulders, M. Worring, S. Santini, A. Gupta, and R. Jain.
Content-based image retrieval at the end of the early years.
IEEE Transactions on Pattern Analysis and Machine Intelli-
gence, 22(12):1349–1380, 2000.
[5] O. Mich, R. Brunelli, and C. Modena. A survey on video
indexing. Journal of Visual Communications and Image Rep-
resentation, 10:78–112, 1999.
[6] E. Izquierdo, J. Casas, R. Leonardi, P. Migliorati,
N. O’Connor, I. Kompatsiaris, and M. G. Strintzis. Advanced
Content-Based Semantic Scene Analysis and Information Re-
trieval: The Schema Project. In Proc. Workshop on Image
Analysis For Multimedia Interactive Services, London, UK,
2003.
[7] V. Mezaris, I. Kompatsiaris, and M. G. Strintzis. Still Im-
age Segmentation Tools for Object-based Multimedia Applica-
tions. International Journal of Pattern Recognition and Artifi-
cial Intelligence, 18(4):701–725, June 2004.
[8] MPEG-7 eXperimentation Model
http://www.lis.ei.tum.de/research/bv/topics/mmdb/
e mpeg7.html
[9] V. Kobla, D. Doermann, and K. Lin. Archiving, indexing, and
retrieval of video in the compressed domain. In Proc. SPIE
Conference on Multimedia Storage and Archiving Systems, vol.
2916, pages 78–89, 1996.
[10] M. Doerr. The CIDOC-CRM An Ontological Approach to
Semantic Interoperabilityof Metadata. AI Magazine, 24(3):75–
92, Fall 2003.
6
... These being carried out through the visual digitized form and the information are being streamlined. It is always essential to provide structured and annotated information in a proper way in order to be accessible and retrievable [2,11]. Some of the initiatives happen where the textile or handicraft designs are being preserved in the form of digital images. ...
Conference Paper
Digital preservation has now become a key activity in every library and information center due to proliferation of Information and Communication Technologies (ICT) and advancement of Internet technologies. The availability of Open Source Software, training at desired time and market demand has revolutionized the creation of digital libraries. Digitization and preservation of readily available resources are common practice but creating digital library of cultural heritage and handicraft materials is still in the novice stage. This paper discusses the creation of digital library of traditional handicraft designs originated from Himachal Pradesh. This paper discusses the tools and techniques of creating a digital library of the handicraft design, which shall be beneficial for the users and fashion designers to look at the creativity of the past, which are still alive in the form of handicraft.
Article
Full-text available
This paper presents the methodology that has been successfully employed over the past 7 years by an interdisciplinary team to create the CIDOC Conceptual Reference Model (CRM), a high-level ontology to enable information integration for cultural heritage data and their correlation with library and archive information. The CIDOC CRM is now in the process to become an ISO standard. This paper justifies in detail the methodology and design by functional requirements and gives examples of its contents. The CIDOC CRM analyses the common conceptualizations behind data and metadata structures to support data transformation, mediation and merging. It is argued that such ontologies are property-centric, in contrast to terminological systems, and should be built with different methodologies. It is demonstrated that ontological and epistemological arguments are equally important for an effective design, in particular when dealing with knowledge from the past in any domain. It is assumed that the presented methodology and the upper level of the ontology are applicable in a far wider domain.
Article
Full-text available
The paper presents a review of 200 references in content-based image retrieval. The paper starts with discussing the working conditions of content-based retrieval: patterns of use, types of pictures, the role of semantics, and the sensory gap. Subsequent sections discuss computational steps for image retrieval systems. Step one of the review is image processing for retrieval sorted by color, texture, and local geometry. Features for retrieval are discussed next, sorted by: accumulative and global features, salient points, object and shape features, signs, and structural combinations thereof. Similarity of pictures and objects in pictures is reviewed for each of the feature types, in close connection to the types and means of feedback the user of the systems is capable of giving by interaction. We briefly discuss aspects of system engineering: databases, system architecture, and evaluation. In the concluding section, we present our view on: the driving force of the field, the heritage from computer vision, the influence on computer vision, the role of similarity and of interaction, the need for databases, the problem of evaluation, and the role of the semantic gap.
Article
Full-text available
The aim of the SCHEMA Network of Excellence is to bring together a critical mass of universities, research centers, industrial partners and end users, in order to design a reference system for content-based semantic scene analysis, interpretation and understanding. Relevant research areas include: content-based multimedia analysis and automatic annotation of semantic multimedia content, combined textual and multimedia information retrieval, semantic -web, MPEG-7 and MPEG-21 standards, user interfaces and human factors. In this paper, recent advances in content-based analysis, indexing and retrieval of digital media within the SCHEMA Network are presented. These advances will be integrated in the SCHEMA module-based, expandable reference system.
Article
Full-text available
Presents a review of 200 references in content-based image retrieval. The paper starts with discussing the working conditions of content-based retrieval: patterns of use, types of pictures, the role of semantics, and the sensory gap. Subsequent sections discuss computational steps for image retrieval systems. Step one of the review is image processing for retrieval sorted by color, texture, and local geometry. Features for retrieval are discussed next, sorted by: accumulative and global features, salient points, object and shape features, signs, and structural combinations thereof. Similarity of pictures and objects in pictures is reviewed for each of the feature types, in close connection to the types and means of feedback the user of the systems is capable of giving by interaction. We briefly discuss aspects of system engineering: databases, system architecture, and evaluation. In the concluding section, we present our view on: the driving force of the field, the heritage from computer vision, the influence on computer vision, the role of similarity and of interaction, the need for databases, the problem of evaluation, and the role of the semantic gap
Article
To support the sharing and reuse of formally represented knowledge among AI systems, it is useful to define the common vocabulary in which shared knowledge is represented. A specification of a representational vocabulary for a shared domain of discourse—definitions of classes, relations, functions, and other objects—is called an ontology. This paper describes a mechanism for defining ontologies that are portable over representation systems. Definitions written in a standard format for predicate calculus are translated by a system called Ontolingua into specialized representations, including frame-based systems as well as relational languages. This allows researchers to share and reuse ontologies, while retaining the computational benefits of specialized implementations.We discuss how the translation approach to portability addresses several technical problems. One problem is how to accommodate the stylistic and organizational differences among representations while preserving declarative content. Another is how to translate from a very expressive language into restricted languages, remaining system-independent while preserving the computational efficiency of implemented systems. We describe how these problems are addressed by basing Ontolingua itself on an ontology of domain-independent, representational idioms.
Article
In this paper, a color image segmentation al- gorithm and an approach to large-format image segmen- tation are presented, both focused on breaking down im- ages to semantic objects for object-based multimedia ap- plications. The proposed color image segmentation algo- rithm performs the segmentation in the combined intensity- texture-position feature space in order to produce con- nected regions that correspond to the real-life objects shown in the image. A preprocessing stage of conditional image fil- tering and a modified K-Means-with-connectivity-constraint pixel classification algorithm are used to allow for seamless integration of the different pixel features. Unsupervised op- eration of the segmentation algorithm is enabled by means of an initial clustering procedure. The large-format image seg- mentation scheme employs the aforementioned segmenta- tion algorithm, providing an elegant framework for the fast segmentation of relatively large images. In this framework, the segmentation algorithm is applied to reduced versions of the original images, in order to speed-up the completion of the segmentation, resulting in a coarse-grained segmen- tation mask. The final fine-grained segmentation mask is produced with partial reclassification of the pixels of the original image to the already formed regions, using a Bayes classifier. As shown by experimental evaluation, this novel scheme provides fast segmentation with high perceptual seg- mentation quality.