[Insight-users] Help : Error Using Discrete Guassian Filter: Effective Resolution

[Luis Ibanez]

Hi Fernando,

Thanks for the suggestion.

That's actually the right way to approach the notion
of image resolution, since pixel spacing is less
relevant once that the image has been blurred.

This notion of FWHM could be a good candidate to
be added to the DataDictionary that has been proposed
as an addition to the itkImage data structure. It
will reflect the effective resolution of the data
contained in the image.

Curiosly, people in astronomy and microscopy pay more
attention to this issues than we do in the medical
imaging community (shame on us).

You may want to point this out in the open forum
of the project involved with the DataDictionary.

Please go to the page:

http://caddlab.rad.unc.edu/itkProjects/indexproject.php?id=27

and look for the open fourm.

You can add topics to the forum as a guest.
In that way your suggestion will not be lost.

You can visit this page frequently to track any
progress in the topics.



   Thanks


    Luis


----------------------------------

Fernando Rannou wrote:
> Hi Luis,
> a small note on the side. When one is working with
> medical images, one usually uses Gaussian filtering
> in the context of image resolution which in many cases
> is measured as a FWHM. Also, in many cases, the image
> to be blurred already has an intrinsic resolution and the
> amount of "actual" blurring applied by the filter depends
> on this. Imaging people often think on terms of FWHM
> and not variance. Anyway, I think it would be nice to
> incluse some of these concepts in ITK.
> 
> Fernando Rannou
> 
> Luis Ibanez wrote:
> 
>> Hi Neha,
>>
>> The DiscreteGaussianImageFilter is described in the
>> software guide in Section 5.5.1, pdf-pages 110 to 112.
>>
>> This filter performs smoothing by convolving the
>> image with a Gaussian kernel. The filter takes
>> advantage of the separability of the Gaussian.
>>
>> The Gaussian kernel is approximated to a level of
>> tolerance set by the user. The lower the tolerance,
>> the better is the approximation, but the larger in
>> size is the kernel.
>>
>> The MaximumKernelSize imposes a limit to the number
>> of pixels in the kernel. You may want to use values
>> in the range of 5 to 10.  Figure 5.14 in the Software
>> Guide will make this clearer for you.  The maximum
>> kernel size should be consistent with the value set
>> for Sigma in the Gaussian.
>>
>> GaussianVariance (the squared value of Sigma) should
>> normaly be in the range of 2 to 10. Note that this
>> particular filter is not taking the spacing into
>> account, so all values are measured in pixels.
>>
>> This shouldn't be a problem, Since you are reading
>> your image from a PNG file, and png files do not
>> contain information about pixel spacing.
>>
>>
>> An example on the use of this filter is
>> available in
>>
>> Insight/Examples/Filtering/DiscreteGaussianImageFilter.cxx
>>
>> This is exactly the same code illustrated in the
>> software guide.
>>
>>
>> Please let us know if you have further questions.
>>
>>
>>   Thanks
>>
>>
>>     Luis
>>
>>
>>
>> -------------------
>>
>> Neha D wrote:
>>
>>> Hi All,
>>>
>>> I am trying to use Discrete Gaussian Filter. I want to use filter on 
>>> png file of unsigned char. What value of maxKernelWidth should I use. 
>>> With values 0.2 , 0.1, I am getting error.
>>>
>>> If anybody is using this filter, please let me know the appropriate 
>>> values for gaussianVariance ,  unsigned int maxKernelWidth .
>>>
>>> Thanks,
>>>
>>>  
>>>
>>
>>
>> _______________________________________________
>> Insight-users mailing list
>> Insight-users@public.kitware.com
>> http://public.kitware.com/mailman/listinfo/insight-users
> 
> 
> 
> 
>