[email protected] chosun univ. the study on the semantic image retrieval using the cognitive...

20
[email protected] CHOSUN UNIV. The Study on the Semantic Image Retrieval Using the Cognitive Spatial Relationships in the Semantic Web Hyunjang Kong,Myunggwun Hwang,Kwansang Na,Pankoo Kim

Upload: aldous-mcbride

Post on 18-Dec-2015

220 views

Category:

Documents


4 download

TRANSCRIPT

[email protected]

CHOSUN UNIV.

The Study on the Semantic Image Retrieval Using the Cognitive Spatial Relationships in the Semantic Web

Hyunjang Kong,Myunggwun Hwang,Kwansang Na,Pankoo Kim

CHOSUN UNIV.

[email protected]

Contents

• Introduction• Related Works• Our Approach• Test and Experimental Results• Evaluations• Conclusion

CHOSUN UNIV.

[email protected]

Introduction

• Huge number of data in the web• Image data is rapidly increasing• Object Based Spatial Relationships

VS Cognitive Spatial Relationships• Building a Spatial Relationships

Ontology

CHOSUN UNIV.

[email protected]

Related Works

• Information Retrieval System– Keyword Matching – Very important technique on the web

environment– Process the various information items

• Text Documents, Images, Sounds and etc.

– Generally, accuracy is low

CHOSUN UNIV.

[email protected]

Related Works

• Ontology based Image Retrieval– Try to solve the heterogeneous between

the terminologies– Need the extra works

• Creating and Maintaining the ontologies

– It is still unsuitable for the image retrieval system• Because it doesn’t consider the features of

the images

CHOSUN UNIV.

[email protected]

Related Works

• The Spatial Description Logic– Region Connection Calculus : RCC-8– Spatial representation is regular subsets of the

topological space– Elementary binary relationships between the

regions• PO, NTPP, TPP, EQ, TPP-1, NTPP-1, EC, DC

X Y X Y YX XY

X Y X Y Y XXY

DC(X, Y) EC(X, Y) TPP(X, Y) TPP- 1(X ,Y)

PO(X, Y) EQ(X, Y) NTPP(X, Y) NTPP- 1(X, Y)

CHOSUN UNIV.

[email protected]

Our Approach

• Background Knowledge of the Cognitive Spatial Relationships

CHOSUN UNIV.

[email protected]

Our Approach

• Building Process of the Spatial Relationships Ontology– Defining the Cognitive Spatial

Relationships– User Research– Using WordNet and Dictionary– OWL Representation

CHOSUN UNIV.

[email protected]

A B

A B

A B

BA

BA

AB

AB

A(B)

C(A,B)

DC(A,B)

PO(A,B)

TPP(A,B)

NTPP(A,B)

TPP- 1(A,B)

NTPP- 1(A,B)

EQ(A,B)

A B

A B

A B

C(A,B)

DC(A,B)

PO(A,B)

RCC- 8

Cognitive Spatial Relationships defined

in our study

Our Approach

• Definition of the Cognitive Spatial Relationships

CHOSUN UNIV.

[email protected]

Our Approach

• User Research– 200 images– 10 people– Clustering the spatial words

CHOSUN UNIV.

[email protected]

Our Approach

• Architecture of the spatial ontology– Upper level– Basic spatial

words level– Instance level

Cognitives_r

connect partof disconnect

verb

v_c v_p v_d

proposition

p_c p_p p_d

kiss lie run swim ride look jump fall

Bussosculatexxxx

Restxxxx

Speed

hurryzipxxxx

Gomovetravelxxxx

Sitdrivexxxx

Frontfaceseexxxx

Leapboundspringxxxx

Pursuenearxxxx

onacrossthroughalongxxxx

atin

aroundroundxxxx

overunderbesidenearxxxx

Upper_Level Cognitive Spatial Relationships

Spatial VerbsBased on the Research

Spatial Verbs Based on the WordNetSpatial Propositions Based on

the OXFORD Dictionary

CHOSUN UNIV.

[email protected]

Our Approach

• WordNet and Dictionary Matching

Cognitive spatial relationships

Research words WordNet matching words

connect Attach Connect, link, tie, link up, fasten, touch, adjoin, meet, contact

connect Kiss Buss, osculate

disconnect Chase Chase after, trail, tail, tag, give chase, god, go after, pursue, follow

disconnect Jump Leap, bound, spring

partof Float Drift, be adrift, blow, swim, transport

partof Hide Conceal, shroud, enshroud, cover, obscure, blot out, obliterate, veil

The spatial propositions based on OXFORD Dictionary

connect On, along, across, through

disconnect Over, under, above, below, by, beside, near, before, behind

partof At, in, around, round

CHOSUN UNIV.

[email protected]

Our Approach

• OWL Representation

CHOSUN UNIV.

[email protected]

Test and Experimental Results

• System Architecture– Contents provider interface– Ontology part– User interface

CHOSUN UNIV.

[email protected]

Test and Experimental Results

• Test Environment– Queries

1. Only one word query – e.g. swan2. Two words query – e.g. swan and lake3. Query containing the spatial relationships –

e.g. swan in the lake4. Natural Language query containing the

spatial verbs – e.g. swimming swan5. Natural Languages query containing the

spatial verbs and proposition – e.g. swan swims in the lake

CHOSUN UNIV.

[email protected]

Test and Experimental Results

• Accuracy Measurement

• Experimental Results

Accuracy = All images searched throughout the system

Correct images matched with the query

CHOSUN UNIV.

[email protected]

Evaluations

CHOSUN UNIV.

[email protected]

Evaluations

CHOSUN UNIV.

[email protected]

Conclusion and Future Works

• Definition of the Cognitive Spatial Relationships

• Applying them to the Image Retrieval System

• Still have Limitation : Semi-Automatic• Our study presents the vision of the

semantic image retrieval and natural language query processing

CHOSUN UNIV.

[email protected]

• Does Anyone Have Any Questions?