New filters/actions for doing FT-based pattern removal can found HERE.
Go there instead of following the tutorial in this post. The process has been greatly improved and semi-automated since I made this post.
Some applications offer FFT/IFFT-filters that can be used in a tedious process to remove repeating patterns (like raster patterns). I wish Photoshop would become the first application to automate this process.
It would work like a filter that – as if by magic – removes repeating patterns in the image:
This is done by performing a “Fast Fourier Transform” (FFT) of the image, splitting it into frequency amplitude and phase (storing it as separate channels). Repeating patterns (frequencies) stand out clearly and can be suppressed by the filter. An Inverse FFT is then performed to convert back to a normal image (now with the repeating patterns removed or reduced).
Note that this only works on a single channel, so normally you would just bother to do this on the Luminosity component. Filtering each RGB-channel separately should be considered. I will leave that up to you Adobe… 😉
For the user, this could be a simple one-click operation, or better yet a slider that affects how aggressively the filter removes frequencies.
Read more to see how to do this in Photoshop manually step by step.
How to remove repeating patterns using FFT
Then we start off with an image that suffers from an evenly repeating pattern (like this paper texture or a raster pattern):
We then run the FFT-filter and end up with a whole lot of seemingly random noise.
The interesting part is the red channel that shows the amplitude of our frequencies:
In a normal photograph we would expect only to see the bright star in the middle, but here we also have lots of small bright star-shapes scattered across our frequencies. Those little stars make up the strong repeating patterns (frequencies) in the image
We remove them using a black brush. This part is tedious but I believe it can be automated and thus making the whole process one simple operation for the user. You have no preview of the result while doing this, so just cover up as much as you can:
When done, run the IFFT-filter to see the result. You might have to go back and forth a couple of times:
I call this amazing! I almost never encounter such images, but I still want this as a filter!
The output is grayscale, so for color images blend the result with the original using Luminosity blend mode (or perhaps run it once per channel?).