SIGIR'98 Tutorial T1
SIGIR'98 logo

21st Annual International ACM SIGIR Conference on
Research and Development in Information Retrieval

Melbourne, Australia

August 24 - 28, 1998

TUTORIAL T1

Multimedia Information Retrieval



Presenter

Norbert Fuhr, University of Dortmund, Germany

Time

Monday 24 August, 9:00am--12:30pm.

Description

The aim of this tutorial is to give a survey on the state of the art in multi-media IR. The focus will be on indexing and retrieval methods for multimedia retrieval, whereas system-oriented aspects will not be addressed.

The tutorial will consist of three major parts, dealing with media, their indexing and their retrieval:

  • In the media part, physical properties of text, audio, images and video will be described briefly. Then the different possible views on each media object are discussed, namely physical, structural, spatial (images and video only), temporal (audio and video only) and perceptive (for images, audio and video). Finally, approaches for modelling the logical structure of multimedia documents are presented.
  • In the indexing part, indexing methods for speech, images and video are described. For speech indexing and retrieval, approaches combining speech recognition with classical text retrieval methods are presented. In image indexing, most approaches are feature-based (e.g. color, texture, contour). Semantic-based approaches use e.g. feature recognition in combination with a rule-based approach for inferring basic semantic concepts. For videos, feature-based approaches deal with the detection of shot boundaries and the creation of representative images for a shot. In restricted domains, also the temporal and spatial structure of videos can be considered.
  • The retrieval part will focus on logic-based approaches, where multimedia IR requires predicate logic for modelling terminological, spatial and temporal relationships. Following this concept, two approaches using terminological logic and probabilistic Datalog will be introduced. Since multimedia documents are structured documents, too, methods for retrieval of structured documents will be considered, presenting two models based on conditional probabilities and four-valued logic, respectively.
  • Audience

    Attendees should have basic knowledge in IR, but knowledge in multimedia systems is not required.

    Biography of presenter

    Norbert Fuhr is Professor in the Computer Science Department of the University of Dortmund, Germany. He is well-known for his theoretical and experimental work on probabilistic indexing and retrieval models. His current research interests are the integration of IR and database systems, networked IR systems, digital libraries and multimedia IR. Concerning the latter, his research group was participating in the joint European ESPRIT project FERMI (Formalization and Evaluation of Multimedia Information Retrieval), where logic-based models for multimedia retrieval were developed. He has given several tutorials at previous SIGIR conferences.

    More information

    The tutorial is an excerpt of the course on multimedia information systems held at the University of Dortmund.

    Cost

    The charge for registration is $A150 per tutorial. Registrants will receive a copy of the notes for the tutorial, and morning/afternoon tea. All tutorials are offered on an only-if-demand-warrants basis; and full refunds will be given for tutorials that are cancelled because of low enrolments. Tutorial notes will also be available for sale on an individual basis at the conference registration desk.
    sigir98@cs.mu.oz.au,
    20 April 1998.