Re: Connectivity used in "Analyze Particle ..." and "Wand (tracing) Tool"

Posted by karo03 on
URL: http://imagej.273.s1.nabble.com/Delaunay-update-tp3694375p3694380.html

Thanks for all the hints, a solution, hopefully not a coding error, is  
for the mentioned image example the application of Process->Binary-
 >Watershed . The thin skeleton lines (8-conn) of the background are  
extended by the operation to 4-connected ones.

After that (without Thresholding) the Wand selects each object nicely  
and Analyze Particle finds (after Thresholding) 19 objects (see http://picasaweb.google.com/lh/photo/3LPKPMUpPUe0iUoxSa7hsQ)
. Gabriel, the 20th under 4-conn. is a one pixel object at the right  
side which is 8-connected to another object.

Regards
Karsten

Am 27.11.2008 um 18:42 schrieb Karsten Rodenacker:

> Gabriel, thank you for the hint. Very helpful! I will see to prepare  
> a macro Wand 2 to test the BinaryReconstruction4/8.
>
> Michael, In fact I did not threshold the image! It is a result of  
> the Delaunay/Voronoi plugin from Johannes (Dscho?) after some  
> broohaha with foreground and background color switches!. However,  
> ImageJ1.42c (daily build) selects particels in the described way! I  
> think I should preserve that example for tests!
>
> Meanwhile I see and remind me to earlier findings that wand selects  
> the next object to the right even if the selection point is not  
> inside the object. This is perhaps an explaination for the queer  
> selection on the right side!
>
> Thanks again for the quick answers
>
> Karsten
>
> Am 27.11.2008 um 18:11 schrieb Gabriel Landini:
>
>> On Thursday 27 November 2008 16:57:39 Karsten Rodenacker wrote:
>>> Hi, Image analysis and -J experts,
>>>
>>> I have an example were I don't fully understand the routines  
>>> "Analyze
>>> Particle ..." and "Wand (tracing) Tool"!
>>>
>>> The image http://picasaweb.google.com/lh/photo/ 
>>> 8Yo_5vJ1zLKZGwP22oDgsw
>>> delivers under "Analyze Particle ..." three particles (white is
>>> foreground). That is understandable, seemingly eight-connected
>>> neighborhood (of foreground) is assumed. Possibly four-connectedness
>>> could be helpful in such cases with thin lines dividing the  
>>> particles.
>>
>> I think that your image has 4-conn particles. If you run Particles4  
>> plugin with the label option you will get 20 particles.
>> If you created this with a watershed/skeletonisation, then to make  
>> it easier, you could use a 4-conneected watershed (which should  
>> give 8-connected particles).
>>
>> If you analyse the image with the Particles8 plugin you get only 3  
>> particles.
>>
>>> But now, trying to overcome this problem by selecting the particles
>>> with "Wand (tracing) Tool", possibly in a macro, I run into
>>> difficulties. Clicking into particles on the left side of the image
>>> selection is okay, Wand is using 4-connectedness! Clicking into  
>>> one of
>>> the two right-most particles seemingly wand is selecting background!
>>>
>>> Is there any way to select one of these two right-most particles?
>>
>> That is a job for the BinaryReconstruct4 plugin in the Morphology  
>> collection. Now this is not compiled into a class, because I have  
>> not had time to test it thoroughly. If you test it, please send me  
>> some feedback.
>>
>> You could get the XStart YStart points with Particles4, then create  
>> a new image with those pixels set for the particles you want to  
>> select and BinaryReconstruct4 with the original as mask and the new  
>> image as seed.
>>
>> I hope it helps.
>>
>> Gabriel
>