An Ontology for Qualitative Description of Images

1
An Ontology for Qualitative Description of Images

• Zoe Falomir, Ernesto Jiménez-Ruiz,
• Lledó Museros, M. Teresa Escrig
• Cognition for Robotics Research (C4R2)
• Temporal Knowledge Base Group (TKBG)
• University Jaume I, Castellón (SPAIN)

2
Motivation (I)

• Our group is applying Freksas Double Cross
  Orientation model to robotic navigation indoors.
• Our robots use a laser sensor to find the
  landmarks of a room which are its corners and the
  corners of the obstacles inside the room.
• Problem sometimes a robot tries to localize
  itself inside a room and the geometry of the
  detected landmarks and its relative situation wrt
  the other landmarks is not enough to solve
  ambiguous situations.
• Solution to describe visually the landmarks of
  the room in order to differentiate easily between
  them.

C1
C2
3
Motivation (II)

• Our approach describes qualitatively any image,
  by describing
• the visual features (shape and colour) and
• the spatial features (orientation and topology)
• of the objects contained in an image.
• An ontology provides our qualitative description
• A formal representation of the knowledge inside
  the robot
• A standard language to exchange information
  between agents
• New information inferred by the reasoners

4
Index

• 1. Qualitative Description of Images
• 1.1. Approach
• 1.2. Models of Shape, Colour, Topology and
  Orientation
• 1.3. Structure of the Description
• 1.4. A Case of Study
• 2. Ontology
• 2.1. Terminological Knowlege Box (T-Box)
• 2.2. Assertional Knowledge Box (A-Box)
• 3. Results
• 3.1. Approach
• 3.2. New Knowledge Inferred from the Case of
  Study
• 4. Conclusion and Future Work

5
1.1. Approach
1. Qualitative Description of Images
Colour graph-based segmentation
Qualitative Models of Shape, Colour, Topology and
Orientation
Image Processing Algorithms
Qualitative Image Description
6
1.2.Models of Shape, Colour, Topology and
Orientation
1. Qualitative Description of Images
Qualitative Shape of relevant point j ltKEC(j),
A(j) or TC(j), L(j), C(j)gt KEC line-line,
line-curve, curve-line, curve-curve,
curvature-point A very-acute, acute, right,
obtuse, very-obtuse TC very-acute, acute,
semicircular, plane, very-plane L
much-shorter (msh), half-lenght (hl),
quite-shorter (qsh), similar-lenght (sl),
quite-longer (ql), double-lenght (dl),
much-longer (ml) C convex, concave
Fixed Orientation
Qualitative Colour Tags black, dark-grey, grey,
light-grey, white, red, yellow, green, turquoise,
blue, violet
Relative Orientation

• Topology Model
• Disjoint (x,y)
• Touching (x, y)
• Completedly_inside (x, y)
• Container (x, y)
• Neighbours Objects with the same container

7
1.3. Structure of the Description
1. Qualitative Description of Images
Qualitative Image Description
Visual Description (1 .. nRegions)
Spatial Description (1 .. nRegions)
Topology (Region)
Fixed Orientation (Region)
Relative Orientation (Region)
Shape (Region)
Colour (Region)
Containers
Reference Systems
Neighbours
8
1.4. A Case of Study
1. Qualitative Description of Images
9
Index

• 1. Qualitative Description of Images
• 1.1. Approach
• 1.2. Models of Shape, Colour, Topology and
  Orientation
• 1.3. Structure of the Description
• 1.4. A Case of Study
• 2. Ontology
• 2.1. Terminological Knowlege Box (T-Box)
• 2.2. Assertional Knowledge Box (A-Box)
• 3. Results
• 3.1. Approach
• 3.2. New Knowledge Inferred from the Case of
  Study
• 4. Conclusion and Future Work

10
2. Ontology

• Provides our qualitative description with
• A formal and explicit meaning to the qualitative
  labels.
• A standard language to share information between
  agents.
• New information inferred by the reasoners
• Tools
• Ontology language OWL3
• Editor Protégé 4
• Reasoners FacT and Pellet
• Knowledge layers
• Reference Conceptualization
• Contextualized Descriptions
• Ontology Facts ? Assertional Knowledge Box (A-Box)

Terminological Knowlege Box (T-Box)
11
2.1. Terminological Knowlege Box (T-Box)
2. Ontology

• Reference Conceptualization represents knowledge
  which is supposed to be valid for any application.

12
2.1. Terminological Knowlege Box (T-Box)
2. Ontology

• Contextualized Knowledge represents a concrete
  domain which is application oriented.

13
2.2. Assertional Knowledge Box (A-Box)
2. Ontology

• Ontology facts represent the individuals
  extracted from the description of the image.

14
Index

• 1. Qualitative Description of Images
• 1.1. Approach
• 1.2. Models of Shape, Colour, Topology and
  Orientation
• 1.3. Structure of the Description
• 1.4. A Case of Study
• 2. Ontology
• 2.1. Terminological Knowlege Box (T-Box)
• 2.2. Assertional Knowledge Box (A-Box)
• 3. Results
• 3.1. Approach
• 3.2. New Knowledge Inferred from the Case of
  Study
• 4. Conclusion and Future Work

15
3.1. Approach
3. Results
16
3.2. New Knowledge Inferred
3. Results

• Inferences
• Object 0 ? UJI_Lab_Wall
• Objects 4, 6 ? UJI_Lab_Door

17
Index

• 1. Qualitative Description of Images
• 1.1. Approach
• 1.2. Models of Shape, Colour, Topology and
  Orientation
• 1.3. Structure of the Description
• 1.4. A Case of Study
• 2. Ontology
• 2.1. Terminological Knowlege Box (T-Box)
• 2.2. Assertional Knowledge Box (A-Box)
• 3. Results
• 3.1. Approach
• 3.2. New Knowledge Inferred from the Case of
  Study
• 4. Conclusion and Future Work

18
4. Conclusions and Future Work

• Our approach describes qualitatively any image
  using qualitative models of shape, colour,
  topology and orientation.
• The qualitative description obtained is
  represented by an ontology, which provides our
  system with
• A formal representation of the knowledge inside
  the robot
• A standard language to exchange information
  between agents
• New knowledge inferred by the reasoners.
• As future work, we intend to
• Extend our approach to integrate the reasoner
  inside the robot system.
• Extend our ontology to characterize and classify
  more landmarks of the robot environment.

19
1 October 2009 is the first birthday of
A spin-off business of
The mission of C-Robots is to provide a 100
fully autonomous solution to existing machinery
automation.
Our product is an intelligent and artificial
brain for service robotics. It is composed of
specific hardware and an intelligent software
which combines traditional robotic solutions with
the more advanced cognitive solutions to address
specific requirements.
We are looking for partners to apply to European
projects
20
Thank you for your attention

• Suggestions to improve our work?