Posted by
ericbarnhill on
URL: http://imagej.273.s1.nabble.com/Changing-fonts-in-macros-for-the-Series-Labeler-tp5000755p5000788.html
Hi all,
I'm wondering if anyone can give me insight into an effect I am
getting in attempting to implement an algorithm.
The algorithm denotes a forward Laplacian operator as, dropping some
constants,
FFT^-1[(p^2+q^2)FFT(f(x,y))]
and a corresponding inverse Laplacian operator as
FFT^-1[FFT(f(x,y)) / (p^2 + q^2)]
in short, forward and inverse Laplacians are obtained by multiplying
and dividing the FFT by a parabolic mask.
x and y are row and column coordinates, p and q are fourier space
coordinates with origin in the middle of the image.
I was quite interested in this algo because it's not too hard to get
the Laplacian, but the inverse Laplacian? Hard.
However, the inverse is not quite working for me. I get a
"checkerboarding" effect where pixels alternate positive-negative. For
example, -55 might be right next to +55. If I take the Math.(abs) of
the inverse it looks like it might be about right, though if so the
algorithm doesn't work so great -- too much high pass on the way in
and too much low pass on the way back. I can attach an image if that
helps clarify anything.
I am wondering, is the checkerboarding a result of the nature of the
DHT? Might I have more luck using some other kind of DFT, or a
discrete cosine transform?
Thanks for any insights,
Eric
--
The University of Edinburgh is a charitable body, registered in
Scotland, with registration number SC005336.
--
ImageJ mailing list:
http://imagej.nih.gov/ij/list.html