Universität Bielefeld - Sonderforschungsbereich 360
Towards Segmentation from Multiple Cues: Symmetry and Color
Nahum Kiryati
Dept. of Electrical Engineering-Systems
Tel Aviv University, Israel
Monday, July 3rd, 2000, 4 p.m., Hörsaal 9
Segmentation is a notoriously difficult problem. It has
been recognized that cue integration might be the key to better
segmentation algorithms. This paper demonstrates the combined use
of color and symmetry for detecting regions of interest (ROI), using
face detection and wooden artifact detection as working examples.
A functional that unifies color fitness and color-symmetry within
elliptic supports is defined. Using this functional,
the ROI detection problem becomes
a five-dimensional global optimization problem. Exhaustive-search
is inapplicable due its prohibitive computational cost.
An adaptive random search method rapidly converges to the correct
solution. The added value obtained by combining color and symmetry
is demonstrated.
Joint work with Roy Shor.
Anke Weinberger, 2000-06-26