Segmentation of multi-component images

[Noel BONNET]

Dear ImageJ users,

I have now released two plug-ins for the interactive or automatic
segmentation of multiple component images:

* the first one (called Segmentation_3_component_image_514) concerns the
segmentation of two-component images. This segmentation can be done
interactively or automatically. In both cases, the scatterplot of the 2
components (viewed as two single images) is first built. For the interactive
option, several parts of the scatterplot are selected and back-mapping to
the image space is performed.
For the automatic method, the probability density function is estimated by
the Parzen method, for different values of the parameter associated to the
smoothing kernel. Then, the user has to select the value of the parameter
that corresponds to the number of classes he is interested in. Then, the pdf
is partitioned into different classes according to the watersheds approach
originating from mathematical morphology. Finally, after backmapping, the
image space is segmented.

* the second one (called Segmentation_max4_component_image_514) performs the
segmentation of 3- or 4-component images semi-automatically, along the same
lines as the other plug-in described above. In this case, the user must
choose parameters corresponding to 3D or 4D scatterplots with small
dimensions, otherwise the computing time is prohibitively long.

The 2 plug-ins are available at:
http://www.univ-reims.fr/INSERM514/ImageJ

I hope these plug-ins will be found useful. Feedback will be appreciated.

Noel

[Gabriel Landini]

On Tuesday 18 October 2005 10:54, Noel BONNET wrote:
> I have now released two plug-ins for the interactive or automatic
> segmentation of multiple component images:

Noel, this sounds very interesting, but I cannot make it work in my setup.
I load the Clown image, then split into rgb and use the Red and Green as the
input. Then the first dialog comes up (asking for thresholds) and produces
the scatterplots.
The second dialog comes up and I reply "Manual", "Scatterplot logarithmic" and
"3" for the number of classes.
Here is where it seems to hang. A new "Segmented Zone" image appears (all
black" and nothing else.

If I use the Automatic mode, it only produces an output only when I check "Fix
sigma" (but I get 139 modes! rathen than the 3 I set).
What am I doing wrong?

I am using java 1.5.0_04 under suse linux 9.2 and IJ 1.35D.

Cheers,

Gabriel

[Gabriel Landini]

I just realised that the automated procedure works fine.

The problem is with the manual one.

Cheers,
Gabriel

[Noel BONNET]

Hi Gabriel,

For the manual procedure: you have to draw as many ROIs as you chose
classes. After drawing each ROI, you have to press "enter" and validate the
color this class will take in the segmented image.
I hope this will help.

Regards.

Noel

PS: Each plug-in is associated to a tutorial than can be reached following
the  link [MORE ...] at http://www.univ-reims.fr/INSERM514/ImageJ

[Gabriel Landini]

On Tuesday 18 October 2005 15:33, you wrote:
> For the manual procedure: you have to draw as many ROIs as you chose
> classes. After drawing each ROI, you have to press "enter" and validate the
> color this class will take in the segmented image.
> I hope this will help.

I see. Thanks, yes that works. I was waiting for the dialog to appear, but
this happens after one draws the first ROI and presses Enter.
What happens if one overlaps one ROI with another. Does the last one take
priority?

Thanks for the clarification. Really interesting plugin.

Cheers,

Gabriel

[Noel BONNET]

"What happens if one overlaps one ROI with another. Does the last one take
priority?"

YES, it does.
Thanks for the encouragements.

Noel

[Noel BONNET]

On Tue, 18 Oct 2005 16:00:34 +0100, Gabriel Landini <[hidden email]>
wrote:

>On Tuesday 18 October 2005 15:33, you wrote:
>> For the manual procedure: you have to draw as many ROIs as you chose
>> classes. After drawing each ROI, you have to press "enter" and validate the
>> color this class will take in the segmented image.
>> I hope this will help.
>
>I see. Thanks, yes that works. I was waiting for the dialog to appear, but
>this happens after one draws the first ROI and presses Enter.
>What happens if one overlaps one ROI with another. Does the last one take
>priority?

YES, it does.


Noel
>=========================================================================

[way4thesub]

The perfect solution was just developed at Stanford University - GemIdent. This open-source free package segments cells from miscroscopic images and, more generally, can identify objects of interest and do color image segmentation for any number of object classes.

Adam

Noel BONNET wrote

Hi Gabriel,

For the manual procedure: you have to draw as many ROIs as you chose
classes. After drawing each ROI, you have to press "enter" and validate the
color this class will take in the segmented image.
I hope this will help.

Regards.

Noel

PS: Each plug-in is associated to a tutorial than can be reached following
the  link [MORE ...] at http://www.univ-reims.fr/INSERM514/ImageJ