Thread Subject: Centroids of bead image

Hi there,

I´m trying to calculate the centroids of many beads as shown
in the following url:
http://img222.imageshack.us/my.php?image=beadste6.png

This however requires a proper detection of each bead.
Judging by eye one can easily distinguish the beads, but if
I do bwareaopen it always takes for example 3 close beads as
one object. I haven´t found a way to get a more precise
label of the single beads. I tried thresholding the image,
applying strel but nothing did work so far. Any idea is much
appreciated.

Many thanks in advance!

Steffen
 

On Nov 20, 12:25 pm, "Steffen " <rile...@gmail.com> wrote:
 SNIP> trying to calculate the centroids of many beads

Have a look at this tutorial:

http://physics.georgetown.edu/matlab/tutorial.html

There is code linked to that page as well that should help you out.

Good luck!

Dave

Hi Steffen,
The following might get you going to separate out the beads.

I = rgb2gray(imread('beadste6.png'));
I = I(2:end-1,2:end-1); %theres a band of 255 values on the
edges
fun = @(x) graythresh(x(:)); %create a nonlinear filter to
do adaptive threshold
gFilt = nlfilter(I,[5,5],fun);
gFilt = uint8(gFilt*255);
BW = false(size(I));
BW(I > gFilt) = true;
g = graythresh(I);
 
BW = im2bw(I,g) & BW; %clean out the low value garbage

figure, imshow(BW);

s = strel('disk',3);
bwOpen = imopen(BW,s); %open it up a bit

figure, imshow(bwOpen);

Hope this helps
Matt

Thanks for the advices!

Is my understand correct that input of the [m m] matrix for
nlfilter corresponds to the feature size in pixels?
I assume so, as the filtering works best when taking the
average radius of my beads into the matrix. However, when I
apply the strel, I´m wondering why this doesn´t work well
for the above found value. In my particular case its a [7 7]
matrix and a strel disc of 4. Why is it not identical and
thus where is my incorrect thougt?


Cheers,
Steffen

"Steffen " <rileksn@gmail.com> wrote in message <fi1lb1
$ris$1@fred.mathworks.com>...
> Thanks for the advices!
>
> Is my understand correct that input of the [m m] matrix
for
> nlfilter corresponds to the feature size in pixels?
> I assume so, as the filtering works best when taking the
> average radius of my beads into the matrix. However, when
I
> apply the strel, I´m wondering why this doesn´t work well
> for the above found value. In my particular case its a [7
7]
> matrix and a strel disc of 4. Why is it not identical and
> thus where is my incorrect thougt?
>
>
> Cheers,
> Steffen

I often make the strcturing element smaller than the
features I am working with if I need to extract a range of
feature sizes. With a disk strel I usually start at about
1/2 the feature size. So for a radius of 7 a disk strel of
3 or 4 would work well. The problem of working with a
structuring element too close to the average feature size
when doing a morphological opening operation is that you
can end up having eroded completely away about half your
features leaving only your biggest features to be re-
dilated. It all depends on what you want to end up with.

Thanks for the info. Seems as though that I´ve to read some
more about it. But the eroding effect makes sense,
theoretically and proven in practive as well. ;)

Thanks!!

> I often make the strcturing element smaller than the
> features I am working with if I need to extract a range of
> feature sizes. With a disk strel I usually start at about
> 1/2 the feature size. So for a radius of 7 a disk strel of
> 3 or 4 would work well. The problem of working with a
> structuring element too close to the average feature size
> when doing a morphological opening operation is that you
> can end up having eroded completely away about half your
> features leaving only your biggest features to be re-
> dilated. It all depends on what you want to end up with.
>
>