Posted by
Michael Schmid on
Dec 11, 2009; 11:25am
URL: http://imagej.273.s1.nabble.com/Rolling-Ball-tp3690109p3690110.html
Hi Christian,
unless you check "sliding paroboloid", at very low values of the ball
diameter you get a kind of edge detection. Even a ball of radius 1
won't touch the edges and corners of a box from the inside. Radii
less than 1 are interpreted as 1 because this algorithm requires a
radius of at least 1.
Furthermore, the data are smoothed before rolling the ball (to reduce
the impact of outliers), so even if you could set radius=0 you would
subtract the smoothed image from the original. There is an option to
disable smoothing.
Also, this legacy algorithm uses some tricks such as downscaling to
reduce the computing time. Actually these scale the ball in a way
that is not really consistent, so, e.g., when changing the radius
from 10.00 to 10.01 you get a jump in the behavior. Since many people
rely on 'Subtract Background' to behave as it always did, this cannot
be changed any more.
If you check "Sliding Paraboloid", the ball is replaced by a
paraboloid that has the same curvature at the apex as a ball of that
radius; here you can have any value of the radius > 0.0001.
Obviously, the width of the paraboloid can be much more than the
radius if there is a large jump of the image data, especially in a 16-
bit or 32-bit images.
By the way, also the "Sliding Paraboloid" algorithm is an
approximation, it does not really use a paraboloid (which would be
very slow) but it rather slides parabolae in different directions
over the image. Doing the full calculation would be very slow...
For RGB images, one more option that you should care about is
'Separate Colors'. If you don't check it, the operation only works on
the brightness, but leaves the hue and saturation untouched.
Michael
________________________________________________________________
On 11 Dec 2009, at 10:31, Christian Breukers wrote:
> Hello,
>
>
>
> I have an image with fluorescence cells on a dark background.
>
> Why are the cells (signal) still present even if I set the rolling
> ball
> diameter much smaller than my typical cell diameter?
>
>
>
> This happens even when I set the diameter to 1 pixel (my cell is more
> than 4 pixels in diameter).
>
> And how come that you can set the rolling ball diameter to even 0.001
> pixel?
>
>
>
> You would expect that the complete signal will be subtracted
> completely
> resulting in a value of 0, so a complete dark image.
>
>
>
> Regards,
>
> Christian
>
> ---------------------------------------------------
> Faculty of Sciences and Technology
>
> MIRA Institute for Biomedical Technology and Technical Medicine
>
> Department of Medical Cell BioPhysics
> The Netherlands
>
>
>
>