[Insight-users] ThresholdSegmentationLevelSet smoothness

[Oleksandr Dzyubak]

Hi Luis,

a=PropagationScaling.
b=CurvatueScaling.
F(a) = F(PropagationScaling).
K(b)=K(CurvatueScaling).

By that I meant functional dependence on PropagationScaling and 
CurvatueScaling parameters.

Alex

Luis Ibanez wrote:
>
> Hi Oleksandr
>
>    Can you share with us what you
>    are referring to by using F(x) and K(x)  ?
>
>    That will be nice   :-)
>
>
>       Thanks
>
>
>          Luis
>
>
>
> -----------------------------
> Oleksandr Dzyubak wrote:
>> Hi Luis,
>>
>> Keeping PropagationScaling=1.0, I varied CurvatueScaling over the 
>> range [1-1000].
>> As you said, there is a smoothing effect but not as much as I expected
>> and the cylinder is still a cylinder.
>> What approach did you use?
>>
>> a) F(a + b)=1
>> b) F(a) + K(b)=1
>> c) F(a*b)->[0-1]
>> d) F(a) ->[0-1] AND K(b) ->[0-1]
>> e) something else?
>>
>> If b) is the case, then setting PropagationScaling << 1.0 should have
>> a smoothing effect as well (or even stronger?), right? For the cost 
>> of sensitivity loss I believe, right?
>> How did you normalize the parameters
>> that such a large CurvatueScaling range is still acceptable?
>>
>> Concerning the ends your are right. I should have used the padding.
>>
>> Regards,
>>
>> Alex
>>
>> Luis Ibanez wrote:
>>
>>>
>>> Hi Oleksandr,
>>>
>>> In order to have the curvature weight parameter to have a significant
>>> effect on the level set you have to set it up to a value that is around
>>> 100X the one of the propagation weight. These values are relative (they
>>> get normalized inside the filter), so their absolute values do not
>>> matter as much as their relative values. Try pushing it to 200 times
>>> the value of the propagation weight... or even higher.
>>>
>>> If you get a rounded sphere as segmentation...
>>> then you will know that you have gone too far :-)
>>>
>>>
>>> About the first and last slice....
>>> Well, how many empty slices are between the end of the cylinder and
>>> the end of the image ?
>>>
>>> You may have to add some padding slices with the PadImageFilter...
>>>
>>>
>>>   Regards,
>>>
>>>
>>>     Luis
>>>
>>>
>>> ------------------------
>>> Oleksandr Dzyubak wrote:
>>>
>>>>
>>>> Dear Luis and ITK users,
>>>>
>>>> I tried to use the ThresholdSegmentationLevelSet image filter
>>>> to segment an elongated gaussian cylinder. The segmentation went fine
>>>> but I still have couple questions.
>>>>
>>>> 1)
>>>> In the segmented cylinder, its edges are too rough with a bunch of 
>>>> spikes.
>>>> According to the user manual, the CurvatureScaling parameter could 
>>>> be used
>>>> as a smoother but it did not help much even if I vary it over a 
>>>> quite wide range [1-100].
>>>> Any suggestions on how to get the segmented object with smooth edges?
>>>> What other control parameters could be used for that?
>>>>
>>>> 2)
>>>> In the segmented image, the first and last slices of the cylinder 
>>>> are excluded.
>>>> It looks like propagation gets stuck at the ends.
>>>> Any suggestions on how to enforce it to process the whole object?
>>>>
>>>> Thanks,
>>>>
>>>> Alex
>>>> _____________________________________
>>>> 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
>>>>
>>
>>