Localization of Symmetries through Iterative Voting
Int. Conf. on Pattern Reconition, 2004
Circular symmetry is an important perceptual
cue for feature-based representation, fixation, and
description of large-scale dataset.
A novel method based on voting
along the gradient direction is introduced
for inferring the center of mass for objects
demonstrating circular symmetries which are not limited
to convex geometries.
A unique aspect of the technique is in the kernel topography, which is
refined and reoriented iteratively. The technique can
detect nonconvex perceptual circular symmetries,
has an excellent noise immunity, and is shown to be tolerant to
scale perturbation. Applications of this approach to
blobs with incomplete and noisy boundaries, multimedia
scenes, and scientific images are demonstrated.
here to see the full version of the paper in Acrobat format