| Science Forum Index » Image Processing Forum » reconstructing image from Zernike moments... |
|
Page 1 of 1 |
|
| Author |
Message |
| Bob Alvarez... |
 Posted: Fri Jul 03, 2009 6:18 am |
|
|
|
Guest
|
On Jul 1, 8:10 am, Ethan <emon... at (no spam) mac.com> wrote:
[quote:088cd26e66]I am trying to implement the reconstruction of an image from its
Zernike moments
[/quote:088cd26e66]
Why are you trying to do this? My experience with moments in general
is that they are extremely numerically unstable. I even avoid using
them as image features. |
|
|
| Back to top |
|
|
|
| Ethan... |
| Posted: Sat Jul 04, 2009 2:21 am |
|
|
|
Guest
|
On Jul 3, 12:18 pm, Bob Alvarez <ralva... at (no spam) spambob.net> wrote:
[quote:4162b026e8]Why are you trying to do this?
[/quote:4162b026e8]
I am trying to reproduce a template matching algorithm that I read in
a recent paper. Zernike moments are rotationally invariant so the
authors use them to choose final match candidates.
I wanted to see if the method was any good, but I wanted to check my
calculations of the ZMs by reconstructing the image and I have been
banging my head against the wall. That's why I asked these questions;
I wanted to see if I am at least implementing the reconstruction
properly.
In the end, the ZMs are computationally expensive, but I still would
like to know if I am doing things correctly. |
|
|
| Back to top |
|
|
|
| DrBZU... |
| Posted: Sat Jul 04, 2009 11:30 pm |
|
|
|
Guest
|
I used Zernike moments to describe & reconstruct images of impact
craters a few years ago. They were quite sensitive to the choice of
origin and I agree with Bob about numerical issues - esp with high
order moments.
Answers to your questions:
1) I believe you can use any point as the origin of the Zernike
decomposition. The quality of the reconstruction will depend on this
choice and how well the Zernike moments happen to describe/fit the
data using the selected origin. This origin sensitivty is one of the
main drawbacks with using them.
2) Best to inscribe the circle within the image rectangle. The higher
order moments will be quite sensitive to zero padding.
3) In theory, you should not have to normalise the image data for the
decomposition to work. However, you may find that you are
encountering numerical problems
with the higher order moments, in which case its rarely a bad idea to
normalise your data to also be in the range [-1..1].
4) I would expect some mild ringing effects by the time you get to
N=40. But coupled with a loss of contrast it may signify a numerical
problem.
Regards
Jason Dale
www.visionexperts.co.uk |
|
|
| Back to top |
|
|
|
|
