In some late experience, we found some dust in our microscope. We decided to save the background without cells in order to remove dust after the acquisition.
We simply divide the original image by the noisy image, but we are not getting what it should:
1st Uneven illumination is still there
2nd The spots are not totally eliminated
3rd Tying Subtraction instead of division is even worst
Do you have any clue of what is the problem? What I should do?
Well too bad it does not work that’s indeed the right way to do background subtraction.
You use the same illumination parameters (exposure/intensity) for the background and with the sample right ?
It could be that the support of the sample also have an impact on the pixel intensities, is the background image with an empty slide ? or even better with a portion of the sample slide where there is no cell ?
For this point I looked at the pixel intensity for a given dust spot in the background and sample image. And there are not the same intensities at all (ex: 5000 for the sample and 2000 for the background) so if you subtract they dont cancel out and you still see the spot…
That’s why I think that could be the effect of the sample mounting.
Ideally the best solution would be to clean the optical system rather than using some image processing…
If this is not doable, well it depends what you want to do. Segmenting the cell looks still looks doable and dust should not be much of an issue.
To get rid of the uneven illumination, the rolling ball (Process>Subtract Background) could be an option too but is maybe more discutable for quantitative analysis, but fine to derive a mask.
I notice that the images looks quite noisy like if there was some grains/salt and pepper noise.
Do you use a long acquisition time ? It’s on a widefield microscope right ?
Hello mmvpgs, (How do you pronounce that?)
I used polynomial shading corrector (an available plugin) set at 2,2,2 on both images then auto brightness/contrast again to both and finely subtracted as previously.
Seems to work alright to me but you are the judge.Result of poly.tif (6.0 MB)
you need a dark frame (subtraction) correction before applying the flat field (division), otherwise you will have an “overcorrection”. That’s because zero intensity does not correspond to a pixel value of zero but some positive value (both due to dark current and an intentional offset in the electronics):
(when working with floating-point accuracy, you can ignore the ‘m’ or ‘Gain’)
When using different gain and/or exposure for the image and flat-field image, each of them should have its own dark frame.
Make sure the flat field has enough intensity; it should not be noisy. You should have floating-point output for the division; otherwise the quantization of pixel values to integers will cause artifacts or make the result completely unusable. If desired, you may set a suitable display range and convert to 16 bits at the end.