Article,
Fast estimation of slopes of linear and quasi-linear structures in noisy background, using Fourier methods
Pattern Recognition (PR), 40 (2): 563--577 (2007)
DOI: 10.1016/j.patcog.2006.04.014
Abstract
We present a fast algorithm for slope detection on gray scale images,
based on 2D Fourier transform and standard filters; this may be used
for line or edge detection. Our approach is based on the calculation
of “energy” per direction of the image, thus obtaining the “energy
spectrum on slope” (θ). This exhibits local maxima at the points
where θ equals the slopes of linear or quasi-linear segments within
the image, yet it is not affected by their position within it. The
process thus outlined has certain advantages as regards its efficiency
of linear structure detection, compared to the Radon and Hough transforms.
It was motivated by the study of astrophysical images (solar dynamic
radio spectra) which necessitated the introduction of a method for
fast extraction of “drifting structures”, since they appear as linear
or quasi-linear segments on these spectra.
