 |  |
 |
Media Files
Abstract
Vision is an amazingly powerful source of information. As a consequence, a considerable part of the data we collect comes in the shape of visual material, and multi-media libraries are rapidly filling up with images and video-footage. Indeed, as it is difficult to categorize images unambiguously, retrieval should be based directly on their visual content to be really useful.
In order to communicate, there is a need for content-sentient vision systems that can assist humans by sifting through vast collections of images and winnowing out the large majority of irrelevant ones. However, in order to be useful such systems ought to be highly adaptive. Indeed, what constitutes an ‚interesting‘ or ‚relevant‘ image will vary widely among users, the context and different tasks or applications call for different criteria by which these images need to be judged.
To tackle this problem, CWI researchers are designing search engines that can be taught what images to look for on the basis of examples. The relevance prediction is based on recently developed non-linear methods for dimension-reduction and probabilistic classification models.
In my talk I will elaborote on interfaces to explore (visual)information by visual means and interaction as well as latest trends in scientific research.
Artists / Authors
- Ben A. M. Schouten, Centrum voor Wiskunde en Informatica CWI, Amsterdam › Biography
Date(s)
- October 25, 2002
Organizer
Fraunhofer Institute for Media Communication MARS-Exploratory Media Lab
Contact
redaktion@netzspannung.org
Location
Schloss Birlinghoven, Sankt Augustin, Germany
Submission
, Apr 10, 2003
Category
- Lecture
Keywords
- Topics:
- databank |
- hypermedia |
- Graphical User Interface GUI |
- knowledge spaces