[Insight-users] anisotropic diffusion or Discrete gaussian

[Aviv Hurvitz]

Hi,

If your goal is to compute the gradient image, the standard thing to do is
to first smooth with a Gaussian (a real Gaussian, not diffusion). This is
because the derivative operator itself outputs the slope of the curve
between neighboring pixels, so if you've got any kind of noise in the image,
it will magnify it in the output. When you smooth the image, it suppresses
the noise.

Anyway, that's the text-book approach. Anisotropic smoothing might work
better in some applications, but I don't know if it's ever been used to
compute the gradient.

BTW, note that GradientRecursiveGaussianImageFilter will smooth the image
and then take the gradient for you in one step, so that can be convenient.

- Aviv



On Fri, Jun 6, 2008 at 1:10 PM, Xavier Mellado Esteban <xme at unizar.es>
wrote:

>
>   Hello,
>
>   You should take a look at ITK User's Guide, if you haven't done it yet:
>
>   http://www.itk.org/ItkSoftwareGuide.pdf
>
>   "Section 6.7 Smoothing Filters" discusses a bit the gaussian and the
> anisotric filters.
>
>   Extracted from the user's guide:
>
>  The drawback of image denoising (smoothing) is that it tends to blur away
>> the sharp boundaries
>> in the image that help to distinguish between the larger-scale anatomical
>> structures that one
>> is trying to characterize (which also limits the size of the smoothing
>> kernels in most applications).
>> Even in cases where smoothing does not obliterate boundaries, it tends to
>> distort the fine
>> structure of the image and thereby changes subtle aspects of the
>> anatomical shapes in question.
>>
>
>
>   I have used them a bit and the main difference is what they say.
> Theorically, the gaussian filter will help to denoise and to have smooth
> derivatives, but it will blur the borders of "small" structures on your
> image. On the contrary, the anisotropic will tend to preserve the borders,
> but I do not know if it has any bad influence on the derivatives.
>
>   In any case, the use of one filter or the other will depend a lot on the
> kind of image and which are the structures you want to study. Besides, the
> parameters used in each filter will change substantially the results, so you
> have to tune them for your particular case.
>
>   My understanding is, if your image is not really noisy and you want to
> help the derivatives, you could smooth it a bit with the gaussian (check
> first that your gradient filter do not smooth it again). If your image is
> noisy, try to denoise it with the anisotropic, and then apply the gaussian
> as I told before. My advice is, try to avoid the gaussian as a denoise
> filter unless your structures are really big.
>
>   Wait for other users, probably they can give you better answers. I hope
> this helps you.
>
>                    Regards,
>                               Xavi
>
>
>
>
> Leila Baghdadi <baghdadi at phenogenomics.ca> ha escrito:
>
>  Hi guys
>>
>> does anyone know the difference between the above two and the pros and
>> cons of using each for before creating the image gradient for
>> segmentation
>>
>>
>> thanks
>>
>> Leila
>>
>> _______________________________________________
>> Insight-users mailing list
>> Insight-users at itk.org
>> http://www.itk.org/mailman/listinfo/insight-users
>>
>>
>
>
> _______________________________________________
> Insight-users mailing list
> Insight-users at itk.org
> http://www.itk.org/mailman/listinfo/insight-users
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.itk.org/pipermail/insight-users/attachments/20080606/cf2898a1/attachment.htm>