__/ [Roy Schestowitz] on Thursday 01 September 2005 17:49 \__
>> [pravesh.subramanian@xxxxxxxxx] on Thursday 01 September 2005 17:33
>>
>> Hello group,
>> This is my first message. I have MRI Images and want to do
>> inhomogeneity correction on them.
>> How can I:
>> 1. determine the noise in the images using Histogram?
>
>
> A histogram might help in detecting salt-and-pepper noise in the case of
> MRI images. It is not, however, a robust method for solving such a task. I
> suspect that the dark background and the light ventricles might pose a
> problem. Consider using localised histograms, i.e. histograms which treat
> smaller portions of the images in question.
>
>
>> 2. perform gaussian smoothing on the image? can anyone give me the
>> function for a gaussian kernel?
>
>
> Which programming language?
>
>
>> 3. normalize this image?
>
>
> Ditto. There are some MATLAB implementations in MATLAB Central and VXL <
> http://vxl.sourceforge.net/ > provides C/C++ equivalents.
__/ Pravs added: \__
> ok... I should have mentioned the language!
> It is Interactive Data Language, IDL. I know something about finding
> the noise pixels. I could smooth the image histogram with a hanning
> kernel first and then i could determine the peak in the first 15 % of
> the intensities. This is an algorithm by some author. He then tells to
> fill the noise locations with the mean of the non-noise pixels.
>
> this image is to be smoothed with the gaussian kernel and then
> normalization is done!
>
> please let me know if you know what this means!
>
> Thanks,
> Pravs
Here is what I can gather:
You get a histogram from your MRI image (wrongly I assumed it was a brain
previously). You then separate out only those values in the histogram which
exceed a certain value. That value should separate apart the histogram so
that it has 85% on one side and 15% on the other (some maths involved
here).
If you know the locations where noise is observed (as indicated above), then
cull them out and calculate the mean over the part of the image that is
noise-free. Having got this mean, use it to fill in points (pixels) that
were culled out as they were noise. Now smooth the image (Gaussian). I
still fail to see how it relates to the first part or /uses/ it. You will
need to provide a description that is a little more detailed. Also, please
quote when replying. It is UseNet, not E-mail so you obstruct other readers
from following the thread.
Hope it helps,
Roy
--
Roy S. Schestowitz | "This sig seemed like a good idea at the time..."
http://Schestowitz.com | SuSE Linux | PGP-Key: 74572E8E
4:00am up 8 days 18:31, 4 users, load average: 0.72, 0.57, 0.49
|
|
