[Insight-users] Correction of non-uniform illumination in EM data

[Lassi Paavolainen]

Hi Nick,

On Wed, 7 Oct 2009, Nick Tustison wrote:

> Hi Lassi, 
> 
> In general, inspection of the corners in an image probably does not
> constitute a proper assessment of bias correction algorithms.  Can you
> describe your data a little more and perhaps I can point you to some
> variations you might want to try?

Ok. I explained it badly. This example dataset has its brightest 
illumination in lower left area of the image. Average intensities then go 
lower to the upper part of the image and to the right part of the image. 
Its some sort of Gaussian distribution where the center is in lower left 
part of the image. I measured those average intensities by calculating 
mean of area about 512x512 in five different places (left-top, right-top, 
center, right-bottom, left-bottom).

Even though mask image is said not to be that important, could it be the 
solution for this problem? In this case using Otsu just masks upper right 
part of the image and elsewhere some noisy parts of the data. At least 
this mask image is quite different of what you would expect to get from 
MRI data. I tested bias correction in EM data.

Lassi

> On Wed, Oct 7, 2009 at 8:46 AM, Lassi Paavolainen <lassi.paavolainen at jyu.fi>
> wrote:
>       Hi Nick,
>
>       On Thu, 24 Sep 2009, Nicholas Tustison wrote:
>
>             Hi Lassi,
>
>             You might want to take a look at the N3 algorithm
>             implemented in ITK.  It might provide what you're
>             looking for.
>
>             http://www.midasjournal.org/browse/publication/640
>
>             Good luck,
>             Nick
> 
> 
> I have played around with that now. It looks promising but doesn't do
> its job completely. I have tested it with one dataset where mean
> intensity is around 130 in lower left corner and around 100 in upper
> right corner.
> 
> When I run test program from your publication using shrinkFactor = 4
> and using default iteration number and number of fitting levels
> values, I get image where difference between those areas is about 20.
> Running it five times difference is subtracted to around 10. I have
> changed the number of iterations and fitting levels without making a
> much of a difference in results.
> 
> So it looks really promising but I would like to get the job done by
> running it only once. Of course results cannot be perfect (there is no
> perfect solution in image analysis). Do you have any idea where I
> should look into to make it converge to some sort of optimal result by
> running the filter just once?
> 
> Lassi
> 
>
>       On Sep 24, 2009, at 4:23 AM, Lassi Paavolainen wrote:
>
>             Hi,
>
>             Does anyone know about ITK method to correct
>             non-uniform illumination? There seems to be
>             MRIBiasCorrection class and an example of it
>             in InsightApplications. I'm not sure if this
>             class can help me and anyway it doesn't seem
>             to be very easy to use.
>
>             I have read following technical report
>             (www.sci.utah.edu/~tolga/pubs/UUSCI-2005-008.pdf).
>             One of the authors seems to be Ross Whitaker
>             who is involved in ITK too. Is there ITK
>             implementation of that method?
>
>             Lassi
>
>             --
>             Lassi Paavolainen, M.Sc.
>             Software Engineer
>             BioImageXD (http://www.bioimagexd.net)
>             lassi.paavolainen at jyu.fi
>
>             _____________________________________
>             Powered by www.kitware.com
>
>             Visit other Kitware open-source projects at
>             http://www.kitware.com/opensource/opensource.html
>
>             Please keep messages on-topic and check the
>             ITK FAQ at: http://www.itk.org/Wiki/ITK_FAQ
>
>             Follow this link to subscribe/unsubscribe:
>             http://www.itk.org/mailman/listinfo/insight-users
> 
> 
> 
> --
> Lassi Paavolainen, M.Sc.
> Software Engineer
> BioImageXD (http://www.bioimagexd.net)
> lassi.paavolainen at jyu.fi
> 
> 
> 
>

-- 
Lassi Paavolainen, M.Sc.
Software Engineer
BioImageXD (http://www.bioimagexd.net)
lassi.paavolainen at jyu.fi