Advanced Search

Browse by Discipline

Scientific Societies

E-print Alerts

Add E-prints

E-print Network

  Advanced Search  

Fast Estimation of Invariant Features Sven Siggelkow and Marc Schael

Summary: Fast Estimation of Invariant Features
Sven Siggelkow and Marc Schael
Albert-Ludwigs-Universitat Freiburg
Institut fur Informatik
79085 Freiburg, Germany
Abstract. Invariant features remain unchanged when the data is trans-
formed according to a group action. This property is very helpful in many
applications when the object's position is irrelevant and one is only in-
terested in its intrinsic features. In this paper we discuss features for the
group of Euclidean motion (both 2D and 3D) that are calculated by an
integration over the transformation group. Although of linear complex-
ity, the computation time is too high if either the application demands
a fast response (e.g. image retrieval) or the size of the data set is high
(e.g. 3D tomographic data). We therefore replace the deterministic com-
putation of the features by an estimation, thus obtaining complexity
O(1) for given error bound and certainty. For a typical 3D application
this results in speedup factors of 10 5 -10 6 . Besides the reduction in com-
putation time, the method is less sensitive to numerical problems which
can become a problem of the deterministic method especially for large
data sets.


Source: Albert-Ludwigs-Universität Freiburg, Institut für Informatik,, Lehrstuhls für Mustererkennung und Bildverarbeitung


Collections: Computer Technologies and Information Sciences