FFT Raster Removal
Have you tried scanning something from a newspaper, and realising that the raster dots
from the print process were quite horrible to look at? I've come up with a method on how
to remove them using an FFT Plugin
Update: The free software Gimp now also has an FFT filter that works well.
Original scanned image:
After this you go through a rather abstract sequence of noise identification and removal
within the fourier transformed space. Some of the involved steps will be explained below.
Resulting image for comparison:
- Transform the image into 2D FFT space. This gives you a frequency representation of the image, where
the presence of large shapes appear as bright areas closer to the center, and smaller shapes gives more
stuff towards the border of the image.
- The image itself will typically show up as a bright star exactly in the center of the image. The raster pattern from the newspaper
print shows up as weird dots next to the center. Now you remove the weird "star" in order to not affect the image itself.
- After this you highpass and level correct the FFT data to retain only the stray dots that represent the raster pattern.
- When these stray dots are isolated, you can subtract them from the original FFT data to get rid of the raster pattern frequencies.
- After this, you do the inverse FFT to convert the strange FFT data back into a normal image.
The real chain of events is a little more complex than this. One problem is that a greyscale image will
turn into a color image when you run the FFT filter. This is a necessary step because the FFT'ed data splits up a greyscale
image into "phase" and "amplitude", which need to be kept separately.
Thus, the process can only handle grayscale pictures or separate color channels.
So to denoise a color image (in RGB mode), i had to go through this process three times: One for red, one for green and one for blue.
To make this entire process less painful, i've created this action
for Photoshop. The action requires this FFT Plugin.
Once again i've tried doing the procss in color. This is somewhat troublesome compared to processing a greyscale image, as you have to repeat the process three times. This trick is now included in the action.
Original scanned image:
Blurred to remove rasters (this is what people often do):
Using my new FFT approach:
Credits to Alex Chirokov for writing this very cool plugin.
Website by Joachim Michaelis