Department of Electrical and Computer Engineering
Beckman Institute, University of Illinois at Urbana-Champaign
Montag, 15.07.2002, 16 Uhr c.t., Hörsaal 9
Compared to text databases, image and video
databases are relatively newcomers.
They offer new possibilities and new
challenges. In particular, for images and
video, it is possible to query by example
and similarity in low-level features
(color, texture, shape/structure, motion,
and audio features in the case of video).
We shall present algorithms for improving
the performance of retrieval by relevance
feedback from the user. Our approach is
to consider the problem as two-category
classification: Relevant and irrelevant
classes. We shall also discuss:
The use of unlabeled data in designing the
classification algorithm; and the combined
use of low-level feathres and keywords in
querying. Some experimental results will
be shown.