|
Title: Triplet-of-Gaussians, ToG: A localizer of curved edges in nD digitized
images
Author: P.W. Verbeek and L.J. van
Vliet
in: Proc. of VIP'93, Int. Conf. on Volume Image Processing (Utrecht, June
2-4), Stichting Computer Vision Research SCVR, Utrecht, 1993, 39-42.
Abstract
Zero-crossing of a second-derivative-of-Gaussian filter is a well-known
edge location criterion. For Gaussian filtered (it holds for all low-pass
filters) curved edges the zero crossing of the Laplacian is shifted away
from the center of the osculating circle.
Difference-of-Gaussians (DoG) filters are similar to the Laplacian-of-Gaussian
and also exert an edge shift to larger edge radii. Here we present a class
of Triplet-of-Gaussians (ToG) filters with reduced curvature-based error.
We have studied the location error of curved edges in D-dimensional images
after analog and digital low-pass filtering. Analog low-pass filtering
is unavoidable since the finite aperture of lenses ensures bandlimitation
of the imaged object. Digital low-pass filtering is often applied to suppress
noise.
No postscript file available.
Return to the homepage
of Lucas J. van Vliet
| 