| | |
Summary: SIMBA { Search IMages By Appearance
Sven Siggelkow, Marc Schael, and Hans Burkhardt
Institute for Pattern Recognition and Image Processing,
Computer Science Department,
Albert-Ludwigs-Universitat Freiburg, Germany
sven.siggelkow@informatik.uni-freiburg.de
http://simba.informatik.uni-freiburg.de
Abstract. In this paper we present SIMBA, a content based image re-
trieval system performing queries based on image appearance. We con-
sider absolute object positions irrelevant for image similarity here and
therefore propose to use invariant features. Based on a general construc-
tion method (integration over the transformation group), we derive in-
variant feature histograms that catch di erent cues of image content:
features that are strongly in
uenced by color and textural features that
are robust to illumination changes. By a weighted combination of these
features the user can adapt the similarity measure according to his needs,
thus improving the retrieval results considerably. The feature extraction
does not require any manual interaction, so that it might be used for
fully automatic annotation in heavily
uctuating image databases.
1 Introduction
|
