Prsentation PowerPoint

1
Extracting an Ontology of Portrayable Objects
from WordNet
S. Zinger, C. Millet, B. Mathieu, G.
Grefenstette, P. Hède, P.-A. Moëllic
Atomic Energy Agency of France (CEA) LIC2M
(Multilingual Multimedia Knowledge Engineering
Laboratory) BP6, 18 Route du Panorama 92265,
Fontenay aux Roses, France sveta_zinger_at_yahoo.com
, milletc,mathieub,grefenstetteg,hedep,moellicp_at_
zoe.cea.fr
2
Goalcreation of a large-scale image ontology

• WordNet lexical resourses
• Image collections acquisition through web-based
  image mining

3
Building a large-scale image ontology for object
recognition
4
Building a large-scale image ontology for object
recognition
5
Pruning approach to WordNet
simplifying connections
selecting branches
6
Extraction from top-level ontology of portrayable
objects
7
VIKA (Visual Kataloguer)
8
VIKA (Visual Kataloguer)
9
Composition of queries
10
VIKA (Visual Kataloguer)
11
Web-image search at present
Desired results
12
Future work

• face detection (adaboost learning) to filter
  web-image search results semantically (images of
  objects without people)
• testing VIKA system performances
• automatic cluster classification ignoring
  irrelevant clusters
• introducing new connections to the ontology
  vision principles (scale), co-occurrence rules

13
(No Transcript)
14
Future work
Co-occurrence rules