How to subtract noise from brightfield image


Hello. Here are two brightfield images attached: one is image of my sample, another one is image taken without sample, but with the same illumination. There is strong noise in this background image.
How is it possible to subtract this background from the original image?


Hi @1timur1

Open both images and navigate to “Process” -> “Image Calculator”.


I was trying it by subtracting background from the image. However what I’ve got was not optimal (attached). I still see same dots. No much change. Is there a better way?


I don’t know what you want to enhance. I am not even sure what you identify as background. Maybe you can take a screen-shot and circle your objects by hand?


Dear Timur,

the thing is that noise generally is random. However, in your case it may not be completely random.

That said, you can’t subtract noise, except in cases in which it isn’t random, i.e. if you deal with a temporally stable background.

You really should try to take two images without specimen and subtract them. If the result isn’t zero or near zero, you have a random component in your background and you must use different methods to get rid of it.




Sure. I circled some of those objects. See attached.


Isn’t that the background image? :confused:

If you can circle the object you want to look at in the “brtfld.tif” we can probably come up with a method which does not include a background image. :slight_smile:


I circled those particles (from background) on brtfld.tif (attached). And I want to get rid of those particles on my brghtfld.tif image.


Oh, now I see. Could it be, that you have some “dirt” in your optical system? I just realized that the values in your background image are in the range 400-700 and the brtfld image has values between 800-2000. This means you first have to normalize both images to the same range of values and then subtract them.

Another question, why would you like to get rid of it? I mean, what would you like to measure in your image, maybe the final measurement/segmentation can be done without removing this particles.


Interesting why values in bckgrnd and brghtfld images appeared to be different. I collected both images with the same illumination.
My goal for this images - neat appearance, nothing else. Just wanted to make it look better by subtracting background. And more importantly - to learn for the future how to process it right to correct dirty particles and uneven illumination.


The absorption/reflection of the material is probably different.

Dependent on the size of the particles you can apply filters (gaussian, median, …) of different sizes. Usually you do that to remove noise. If you have bigger particles which decrease the image quality you can segment and filter them or what makes image analysis much simple, improve the acquisition pipeline until you get perfect images.

Uneven illumination can be corrected as follows:
image - gaussianFiltering(sample, large sigma) + mean(image)

Enhancing contrast can be done with CLAHE.

In general image analysis is very image (acquisition system) dependent.