Particle Analysis and Watershed

[Florian Sedlmeir]

Hi,

Now I have started to use the auto-threshold and local-auto-threshold  
plugins for my particles (problem was: the particles are locally  
distinguishable against the background due to their contrast, but  
globally, there is always a long range gradient in grey values which i  
can´t substract sufficiently)
The new problem with local thresholding algorithms is: My particles are  
very close together and their edges are darker then the middle. So there  
seems to be no optimal radius for local thresholding. Usually i get only  
the edges thresholded (of course not completly... )

Question about watershed: After thresholding i have to use the watershed  
tool as the particles are very close together and touch each other  
sometimes. The probem is, that the particles are not totally konvex, they  
have roughly the form of a kidney. So the watershed tool always divides  
the particles in two halfes.  Of course i can correct this manually  
(drawing one line is enough) but that´s not very funny for many particles.  
Does anyone know a solution of this problem?

Thank you and best regards.
Florian

[Johan Henriksson-2]

Florian Sedlmeir wrote:

> Hi,
>
> Now I have started to use the auto-threshold and local-auto-threshold
> plugins for my particles (problem was: the particles are locally
> distinguishable against the background due to their contrast, but
> globally, there is always a long range gradient in grey values which i
> can´t substract sufficiently)
> The new problem with local thresholding algorithms is: My particles
> are very close together and their edges are darker then the middle. So
> there seems to be no optimal radius for local thresholding. Usually i
> get only the edges thresholded (of course not completly... )
>
> Question about watershed: After thresholding i have to use the
> watershed tool as the particles are very close together and touch each
> other sometimes. The probem is, that the particles are not totally
> konvex, they have roughly the form of a kidney. So the watershed tool
> always divides the particles in two halfes.  Of course i can correct
> this manually (drawing one line is enough) but that´s not very funny
> for many particles. Does anyone know a solution of this problem?

pictures always help but I see where this issue is coming from. you
might be able to recombine the particles afterwards with some heuristics
(if two areas are too circular and they have the least color difference
in the "bridge" as opposed to other combinations).

another approach would be with active contours with prior shape information

/Johan


>
> Thank you and best regards.
> Florian

[Michael Schmid]

Hi Florian,

concerning Process>Binary>Watershed: You can have a 'watershed  
segmentation' function of binary images with less sensitivity:

If you have small particles, for better accuracy, select 32-bit EDM  
output in the binary options.
Run Process>Binary>Distance Map on your binary image
Threshold between 1 and 255
Run Process>Binary>Find Maxima with 'Above lower threshold', output  
type 'Segmented particles' and use preview to adjust the tolerance in  
a way that each of your kidney particles gets only one maximum. (use  
'light background' if your distance map has a white background)
With a proper value, you can still split particles that are slightly  
touching.

The Process>Binary>Watershed command does essentially the same, with  
the only difference that it uses a special algorithm to reduce  
quantization noise caused by the finite pixel spacing. This is  
important at low values of the tolerance (say, below 1).


Michael
________________________________________________________________

On 10 Dec 2009, at 19:38, Florian Sedlmeir wrote:

> Hi,
>
> Now I have started to use the auto-threshold and local-auto-
> threshold plugins for my particles (problem was: the particles are  
> locally distinguishable against the background due to their  
> contrast, but globally, there is always a long range gradient in  
> grey values which i can´t substract sufficiently)
> The new problem with local thresholding algorithms is: My particles  
> are very close together and their edges are darker then the middle.  
> So there seems to be no optimal radius for local thresholding.  
> Usually i get only the edges thresholded (of course not completly... )
>
> Question about watershed: After thresholding i have to use the  
> watershed tool as the particles are very close together and touch  
> each other sometimes. The probem is, that the particles are not  
> totally konvex, they have roughly the form of a kidney. So the  
> watershed tool always divides the particles in two halfes.  Of  
> course i can correct this manually (drawing one line is enough) but  
> that´s not very funny for many particles. Does anyone know a  
> solution of this problem?
>
> Thank you and best regards.
> Florian