The microscopy image below is corrupted by an irregular checkerboard-like grid. I attempted running 2D-FFT on the image, but was unable to isolate the frequencies which cause the pattern.
How would I go about removing the interference pattern? The method should preferably generalize to instances where the grid has a different size and angle of twist.
Any help would be greatly appreciated.
Assuming that we are talking about one or a few images (not thousands), and that you have exhausted the possibilities of improving the physical capture.
Work in linear light if possible («raw» not jpeg or anything with an applied gamma function). Do a manual rotation (or affine transform) to make he verticals perfectly vertical and the horizontals perfectly horizontal Manually find he offsets of each row/colum of tiles. See if they are exactly the same size.
Import all of the tiles into a script (my choice would be Matlab YMMV), do an average (or possibly median) to get a «proto-tile» without the local image information.
Generate a new image of that proto-tile placed a the exact same spot as the original tiles. Subtract.
One might hope that this removes much of the gradients. You will probably still have discontinuities at the tile borders though.
That looks to me like a composite, satellite photos stitched together. Additionally, all the individual photos have vignette.
You need to fix this before it happens. Learn about "flat-field correction". That will get rid of the vignette.
It's basically the same approach as given in the other answer, but now you know what the procedure is called, and that you need to take a few steps back to fix this properly.
It's also usually done explicitly, before any other steps like stitching/compositing, and it's done using data you know is "gray". You shouldn't attempt to make a flat-field composite from heavily tinted/biased data, like pictures of the ocean (all blue) or forests (all green, I hope). Pictures of dirt/sand might be as close to neutral as you can feasibly get, but even those need white-balance correction (to adjust for the beige tint). Or you can find a large area covered in snow, which is perfect. FFC is usually done on long exposures. Summing of individual pictures is a digital version of long exposure. You'd get the best results from physical long exposure pictures. It may also pay to defocus the lens, as long as that doesn't affect the vignetting (it may...).
