| | |
Summary: Patch Based Localization of Visual Object Class Instances
Alexandra Teynor and Hans Burkhardt
Universit¨at Freiburg
Chair of Pattern Recognition and Image Processing
79110 Freiburg, Germany
{teynor,burkhardt}@informatik.uni-freiburg.de
Abstract
Huge image databases require the automatic analysis of
image content in order to retrieve information. Especially
the detection and localization of visual object class mem-
bers is an important issue. In this work, we deal with the lo-
calization of visual object class members in a patch based
object recognition framework. In particular, we show how
not only the location and scale of an object can be deter-
mined, but also the orientation, a parameter typically ne-
glected in current localization systems. Our method uses
features computed at Difference of Gaussian interest points
and remembers the orientation of the local patches relative
to the reference object. Using a general Hough transform
like voting scheme, the position and orientation of query ob-
|
