Hi,
On Wed, 3 Feb 2010, Konstantin B wrote:
> Hi, I am new to imagej and in need of help(actually I'm using fiji if this
> makes a difference). I'm trying to separate the blue dots from the green
> clusters in the attached picture and count each. So far I've figured out how
> to split the color, channels and have attempted to blur the clusters
> together by messing with the Gaussian blur and then finding edges/
> sharpening until I get blobs. This works pretty well for whats in the
> center, but the clusters in the periphery are stretched out and I'm not sure
> how to include them, is this going to be possible?.Also, I'm having trouble
> figuring out appropriate parameters in terms of pixel size for the actual
> counting.
As you found out, the variance in contrast is the most serious problem. I
would solve it by duplicating the green channel, doing a maximum filter
with a really large radius (something like 50), then divide the green
channel by that maximum (you need to make a float result, otherwise you
will lose precision).
After that, a simple threshold already brings out almost what you want.
You might need to make a binary from that, applying some morphological
operations (maybe a single erosion alread helps).
The more problematic thing now is how to separate the blue channel from
the green one, as the green intensities are (weakly) visible even in the
blue channel. In my experiments, dividing the green channel by 3 and then
subtracting it from the blue channel helped. Then an automatic threshold
followed by watershed to separate the touching clusters.
NOTE: this was just a quick experiment, and could lead the way to a proper
analysis. Real quantification will require you to have a proper
illumination model (the maximum was just a quick method to guesstimate
that), and a proper separation of the channels _before_ your CCD takes
the image. And you will have to verify that the result makes sense by
comparing with manual counting for a good number of images.
Hth,
Johannes
Locked
. Here it is:
