|
|
|
Subspace Methods in Computer Visionby Horst Bischof,
DI.Dr.
Graz University of Technology
Inst. for Graphics and
Vision
Abstract:
Subspace
methods have become a standard tool in the vision community
to
perform visual learning and recognition. These methods are
based on principles
originally used for statistical pattern
recognition. Visual information is
treated in a
direct---view-based manner. Therefore, these methods are
not
limited by objects' geometric complexity, texture, or surface
markings. This
direct representation and the link to statistical
pattern recognition make
these methods much more suitable for
learning.
In this presentation we will review the basic
ideas of subspace methods for
visual learning and recognition. We
will address both supervised and
unsupervised methods such as
Principal Component Analysis (PCA), Linear
Discriminant Analysis
(LDA), Independent Component Analysis (ICA), and
Canonical
Correlation Analysis (CCA). All the concepts
introduced
throughout the tutorial will be demonstrated on the
tasks such as object
recognition and visual localization of a
mobile robot using
panoramic
images.
| |
SSIP 04 
Registration
