Summary: Feature Selection for Retrieval Purposes
and Hans Burkhardt1
University of Freiburg, Computer Science Department,
79110 Freiburg i.Br., Germany
Abstract. The quality of a retrieval system relies to major part on the
quality of the used features. The features have to be small and compact,
but also discriminative. Feature selection is one way to achieve both goals.
We present a new feature selection method with the focus on retrieval
purposes. The new method is based on the well known Relief algorithm.
The new algorithm is shown to be superior to state-of-the-art methods
both on toy problems and real-life 3D-Shape and image retrieval tasks.
The algorithm is based on the intuition that distances to false detection
has to be enlarged and distances to non-detected positives has to be
Feature Extraction and Feature Selection are important tasks in many fields of
computer vision. Given a retrieval or classification task, the goal is to find a
transform from a possible high dimensional feature space to a low dimensional