[Insight-users] contours extraction

Sat Jun 19 10:33:23 EDT 2004

Hi Yann,

Thanks for your clarifications.
The processing that you suggest makes sense in that context.

However,

            Please note that processing pieces of contours is
a tricky business. It often leads to frustration, depression,
desperation, and in some critical cases it may even lead to
premature and sudden death of the image analyst     :-)

The biggest challenge with contours and edges is to have a
mathematically consistent definition of the elements that you
want to use. In this regard it is very important that you don't
use 4-connectivity or 8-connectivity. If you are decided to walk
the dark path of contour processing, do yourself a favor and start
by using 6-connectivity.

Before you start walking that dark path, you may want to consider
other approaches. For example, instead of displaying the edges
(good and bad ones) you could display the internal regions of cells
as colored regions, and allow an operator to select to contiguous
regions in order to fuse them. Note that this is equivalent to the
process of removing the edge separating these two regions, but it
prevent you from going through the pain of verifying the topological
properties of the contour.

I wonder if you have tried the Watershed algorithm in your cell
images. You may want to give it a try to the program in examples:

   Insight/Examples/Segmentation/
                       WatershedSegmentation1.cxx

Watersheds will give you a very ergonomic environment where
regions are defined at different water levels and then can be
selected by an operator in order to do a supervised refinement.
This will be much easier than dealing with contours.

Josh Cates did a detailed description of this algorithm in the
SoftwareGuide

       http://www.itk.org/ItkSoftwareGuide.pdf

In an interactive environment, you will run the filter once,
and then navigate through the tree structure of regions that
are generated by the filter.  Note that this filter is more
effective if you pre-process your images with any of the edge-
preserving filters described in the SoftwareGuide.

Please let us know if you have further questions,

    Thanks,

      Luis

-------------------
Yann GAVET wrote:

> Hi luis, thank you for your answer.
> 
> I will add some precisions: the detection of the cells , due to the bad 
> quality of the images, is not perfect. So, the operators want to have 
> the capability of deleting and adding borders.
> To implement this, I have to group the border pixels into segments.
> But before doing that, I want to close the borders that have little 
> holes in it and that can easily be seen, at least with a human eye.
> 
> The really final goal is to compute statistics on cells, such as area 
> and morphometry. And I want to be able to correct the drawing of the 
> cells automatically done by the algorithm (cell deletion and addition).
> 
> I hope those explanations are better.
> 
> Thank you again.
> Yann.
> 
> Luis Ibanez wrote:
> 
>>
>> Hi Yann,
>>
>> The processing that you are attempting to implement
>> is completely dependend on the discret topology
>> that you define for your pixels.
>>
>> The first thing that you want to avoid is to use
>> 4-connectedness and 8-connectedness since both
>> of them are flawed digital topologies, and for
>> your application it will be hopeless to attempt
>> a correct definition of what a "node" means.
>>
>> The more consistent way of managing connections
>> in a discrete space in 2D is to use a 6-connected
>> topology.  In that context you could use the typical
>> engineering approach of considering connections
>> along the vertical and horizontal axis, and along
>> *only one* of the diagonals. In this way you get a
>> 6-connected space where the notions of neighbors
>> can be properly defined.
>>
>>
>> You may want to look at the classical literature
>> on this area:
>>
>> A. Rosenfeld, A. Kak
>> Digital picture processing
>> Academic Press Inc. 1982.
>>
>> J. Serra
>> Image Analysis and Mathematical Morphology
>> Academic Press Inc. 1982
>> (You will notice that most of the "formal"
>> Mathematical Morphology is actually done in
>> hexadecimal grids)
>>
>> D. Nogly, M. Schladt
>> Digital Topology on Graphs
>> Computer Vision and Image Understanding,
>> Vol 63, No.2; March; pp 394-396, 1996.
>>
>>
>> Your description was a bit confusing regarding
>> what you already have solved and what you are
>> looking for.
>>
>> It seems that you already have succed to detect
>> the regions separating the cells, and now your
>> are trying to identify the cells themselves.
>>
>> It is not clear why you need to identify the
>> nodes in the contours and why you plan still
>> to remove borders that are "not well detected".
>>
>> Maybe a description of your "real final" goal
>> (not the intermediate difficulties) will be
>> useful for suggesting you appropriate ways
>> of taking advantage of the algorithms available
>> in ITK.
>>
>>
>>   Please let us know,
>>
>>
>>
>>      Regards,
>>
>>
>>
>>          Luis
>>
>>
>>
>>
>> ------------------
>> Yann GAVET wrote:
>>
>>> Hi all.
>>>
>>> I want to perform a contour extraction from cell images.
>>>
>>> My program computes the skeleton of the contours, and I want to get 
>>> the cells (with itkConnectedComponentImagefilter).
>>> I also want to get the borders of the cells in a specific way, so 
>>> that I could remove easily the borders that are not well detected. 
>>> The important thing should be the detection of the "node" points, in 
>>> red in the picture. Then, I could separate the borders of the cells 
>>> into segments and manage segments easily.
>>>
>>> My question is: is there a way in itk to perform such a segmentation 
>>> ? Does anyone know any method or link that could help me ?
>>>
>>> Thank you for your time.
>>>
>>> Yann
>>>
>>> ------------------------------------------------------------------------
>>>
>>> _______________________________________________
>>> 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