Edge localization by MoG filters: Multiple-of-Gaussians

Zero-crossings in a second-derivative-of-Gaussian filtered image is a well-known edge location criterion. Examples are the Laplacian and directional second derivatives such as the second derivative in the gradient direction (SDGD). Derivative operators can easily be implemented as a convolution with...

Full description

Saved in:
Bibliographic Details
Published in:Pattern recognition letters 1994-05, Vol.15 (5), p.485-496
Main Authors: van Vliet, Lucas J., Verbeek, Piet W.
Format: Article
Language:English
Subjects:
Applied sciences
Artificial intelligence
Computer science
control theory
systems
Edge accuracy
Exact sciences and technology
Gaussian filter
Laplacian-of-Gaussian
Multi-dimensional edge detection
Pattern recognition. Digital image processing. Computational geometry
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Zero-crossings in a second-derivative-of-Gaussian filtered image is a well-known edge location criterion. Examples are the Laplacian and directional second derivatives such as the second derivative in the gradient direction (SDGD). Derivative operators can easily be implemented as a convolution with a derivative of a Gaussian. Gaussian filtering displaces the half height isophote towards smaller edge radii (inwards for convex edges and outwards for concave edges). The Difference-of-Gaussians (DoG) filters are similar to the Laplacian-of-Gaussian and exert an edge shift to larger edge radii (outwards). This paper introduces Multiple-of-Gaussians filters with reduced curvature-based error. Using N Gaussians ( N>2) we reduce edge shifts to a fraction (1/(2 N−3)) of the ones produced by a similar Laplacian-of-Gaussian filter.
ISSN:0167-8655
1872-7344
DOI:10.1016/0167-8655(94)90140-6