[ITK Community] Isolating 3d touching regions

[Emiliano Pastorelli]

Hi guys,

thanks, there's quite some material to chew on there, i'll take a look 
at it today in the office. The whole process for me needs to be somehow 
automatized from the beginning to the end to achieve what i need, so the 
user will only have to specify the fibres size and length and no manual 
thresholding.

Thanks, i'll get back to you if I have further problems!

Best,
Emiliano

Il 14/03/14 03:13, Bradley Lowekamp ha scritto:
> Hello,
>
> You may also find this SimpleITK IPython Notebook which used the this suggested watershed approach to separate some coins:
> http://simpleitk.github.io/SimpleITK-Notebooks/32_Watersheds_Segmentation.html
>
> Also SimpleITK from Python may be a useful way to easily explore the many algorithms and infinite number of combinations available from ITK.
>
> Brad
>
> On Mar 13, 2014, at 5:37 PM, Dan Mueller <dan.muel at gmail.com> wrote:
>
>> Hi Emiliano,
>>
>> One possibility might be to use the mathematical watershed algorithm
>> to binary segment the fibres and separate touching labels.
>>
>> See here:
>> http://www.insight-journal.org/browse/publication/92
>>
>> and here:
>> http://www.vincent-net.com/luc/papers/96semstats_morpho_topics.pdf (pg 44)
>>
>> This would essentially reduce your problem to find a good "marker" for
>> each fibre. In your case this should not be too difficult, as the
>> skeleton of each fibre should suffice (i.e. the skeletons will not
>> overlap).
>>
>> Good luck.
>>
>> Cheers, Dan
>>
>> On 13 March 2014 23:16, Emiliano Pastorelli
>> <emiliano.pastorelli at gmail.com> wrote:
>>> Hi All,
>>>
>>> i'm quite new to ITK, and i'm trying to manipulate a 3d volume image
>>> representing some fibres floating in a block of concrete. Due to the
>>> scanning inaccuracies, some of them actually seem to "touch", also when they
>>> wouldn't be touching in the real dataset.
>>>
>>> I tried to separate them in every way, through the Frangi Vesselness
>>> filters, and another bunch of attempts, but i don't seem to find a way to do
>>> it reasonably quickly and smartly.
>>>
>>> http://www.kyb3.org/images/touchingFibres.png
>>>
>>> in the picture, exactly in the center you see an example of how the touching
>>> area usually looks like (the dataset is already divide in labeled regions as
>>> label map as well).
>>> I am pretty sure that some filter can do that, but for example i was taking
>>> a look at the connectedthresholdimagefilter and i couldn't really figure out
>>> if that's what i need, and in that case how i shall use the seeds (my
>>> software shall become flexible enough to deal with really similar in shapes
>>> but different in distribution, alignment and contact points datasets).
>>>
>>> Does anybody have any idea that might spare me another couple of endless
>>> days like the last ones?
>>>
>>> Thanks in advance,
>>> Emiliano
>>> _______________________________________________
>>> Community mailing list
>>> Community at itk.org
>>> http://public.kitware.com/cgi-bin/mailman/listinfo/community
>> _______________________________________________
>> Community mailing list
>> Community at itk.org
>> http://public.kitware.com/cgi-bin/mailman/listinfo/community