Is it possible to filter out noise that is less than 1 pixel in size?

Hi everyone. I have a set of images that look like this up close:

I want to filter out the noise in these images, but this noise seems to be less than 1 pixel in size, so traditional filtering with 3x3 kernels for example don't work here. Is there any way to filter such small noise?

[Rik]

It is not immediately clear to me what the noise is in this case. Also, less than a pixel is not possible in the case of noise, but that might just have been lost in translation.

[Andrea Labudzki]

Essentially these are grid lines that should be uniform and the white openings should be rectangular, but there is some noise which occurs here and causes the problem of the openings to not have a regular shape.

[Rik]

Without a manually corrected image it is not clear to me what the result should be.

[Andrea Labudzki]

The original images look like this. I want the grid lines to be "cleaner" without this noise that is visible on the first image I posted.

Though now that I'm looking at the original image, I think the noise that is unwanted is exaclty the size of one pixel. I've tried erosion and dilation on it but it didn't really do much.

[Rik]

The problem is that your grid lines are distorted. I don't know what you want to do as your next step, but I don't think the 1 pixel deviations are currently your problem.

What kind of analysis did you want to do next? Find out what percentage of each square is black?

[Andrea Labudzki]

Yea i'm aware of this problem though this is not up to me. this image was provided to me and I will be provided with better quality images soon. I am doing analysis to find the width and height of the grid lines and the openings.

Answer 1

[Image Analyst]

There is a fisheye/barrel/pincushion distortion routines: undistortImage, undistortPoints, undistortFisheyeImage.

Once corrected, I'd sum the image vertically and horizontally to get the average vertical and horizontal profiles. then you can threshold those profiles to determine the location of the edges. Then in between those edges you can erase the values.