removing ring artefacts

[Don R. Baker]

Hello All,

We are working on micro-tomography images of bubbles in glass
collected at the Advanced Photon Source.  We are having problems
removing the ring artefacts while still retaining the bubbles because
they appear very similar in many of our images.  We have developed a
series of filtering steps that removes most of the rings, but we still
have some problems with rings, especially near the centre of the image.
I have checked the archives of this mailing group and the plugins page
for ImageJ and cannot find any evident solution to the ring artefact
problem, but yet I know that this problem is a common one so I feel
there must be a solution out there that I have simply missed.

Any help you can offer will be appreciated very much.

Wishing you all the best from Montreal,

Don
--
"Melting rocks today for a better tomorrow . . . "
Don R. Baker, Professor of Geochemistry, Earth and Planetary Sciences,
McGill University 3450 rue University, Montreal, QC Canada H3A 2A7  514-398-7485

[Michael A. Miller]

>>>>> "Don" == Don R Baker <[hidden email]> writes:

    > Hello All, We are working on micro-tomography images of
    > bubbles in glass collected at the Advanced Photon Source.
    > We are having problems removing the ring artefacts while
    > still retaining the bubbles because they appear very
    > similar in many of our images.  We have developed a series
    > of filtering steps that removes most of the rings, but we
    > still have some problems with rings, especially near the
    > centre of the image.

Don,

If you have detectors that are poorly calibrated (normalized),
then you will see ring artifacts.  It may be that your
normalization data is noisy, or that something has changed since
you did the normalization measurement.  Either way, a higher
quality normalization measurement should remove the rings for
you.  If the system has not been changed since you acquired the
original data, it may be simplest to acquire better normalization
data and use that in your reconstruction.

If things have changed, so you cannot reproduce the artifacts in
order to normalize them out, you can do the corrections in the
sinograms (if you have them).  In a sinogram the artifact will be
a line, while the bubbles will look like sinusoids.  So in the
parts of the sinogram that do not have data from the objects (the
bubbles), you should see a uniform background with linear
artifacts.  This will be tricky if the object is large, so that
there aren't un-attenuated projections available in the
sinograms.  If you correct the projections so that the artifacts
are filled, then you can reconstruct the image from the corrected
sinogram and be done.  (This method is similar to a scatter
correction for emission tomography)

Even if you don't have the sinograms, you could forward project
the images to get a sinogram, correct the sinogram and then back
project back to the images.  That should work if the images are
not very noisy, and maybe even if they are...  I'll have to think
about that a bit.

If you must work with the reconstructed images, rather than the
sinograms, you might try using a radial gradient filter centered
on the center of rotation of the imaging system.  That should
identify the rings and exclude the bubbles, as long as they are
not distributed just like the rings that is - from your question,
it sounds like that might be the case.

Mike

--
Michael A. Miller                               [hidden email]
  Imaging Sciences, Department of Radiology, IU School of Medicine