Analyse et indexation d

1
Analysing and indexing graphical documents images
Mickaël COUSTATY L3i University of La
Rochelle thesis supervised by Jean-Marc OGIER
2
Context

• Increasing interest to preserv and freely
  access to historical documents
• Many sources possible libraries, museum,
  archives,
• Many various origins, heterogenous and non
  structured documents
• No a priori knowledge
• Huge documents masses

University of La Rochelle Laboratory of
Informatic, Images and Interactions
2
Mickaël Coustaty
3
Objective

• Every one access to any graphical image from
  anywhere
• Anyone / Anywhere / Anything / Fastly and
  Efficiently

University of La Rochelle Laboratory of
Informatic, Images and Interactions
3
Mickaël Coustaty
4
Indexation steps

• Signatures processing (statistical / structural)
• On all image ?
• On certaines areas ?
• On interest points ?
• On colour ? On shape ? On texture ?
• Couple structurals and statisticals signatures ?
• Indexation and classification
• Which classifier to use ?
• Which metrics to use ?
• Problems of data masses
• clustering
• lattices
• graphs (combinatorial problems)

5
Existing projects

• Google (books, image)
• QBIC (IBM)
• Live search (images et books)

University of La Rochelle Laboratory of
Informatic, Images and Interactions
5
Mickaël Coustaty
6
Potentials ideas

• Statistical description of interests points and
  coding of their spatial organization Rusiñol et
  al.
• Use of a Galois Lattice for classification /
  indexation Guillas et al.
• Identification of interest zones and processing
  of adaptated signatures to this area Journet et
  al. Surapong et al.

University of La Rochelle Laboratory of
Informatic, Images and Interactions
6
Mickaël Coustaty