FFT - is what I tried mathmatically sound?

[juliewood]

I love using the FFT filter, but don't understand enough about the mathematics behind it to analyze exactly what it is doing. I experimented today, and I'm curious if anyone who does understand the math can tell me if my experiment should work consistently, or if it was probably a fluke of this particular image.

I have a very poor-quality image that a customer would like blown up. (It is a poor scan of an already overenlarged inkjet printout from a very low mb camera, but it is the last photo of her dead teenage son and she was in tears, so I agreed to at least try to do something with it.) It is very jpeg-y and noisy, but because it is already enlarged my regular jpeg "fixers" don't recognize the large blocks. My experiment actually did seem to help...

When I ran FFT, there were oh so many stars in the red channel and it was going to take forever to paint them all out. So what I did was:

1) Ran the FFT action until it stopped for me to paint the stars.
2) Selected the Red Channel, then Select All, then created a new layer and pasted.
3) Inverted it.
4) Adjusted levels so I mostly just saw stars.
4) Ran Median at 1.
4) Set it to darken.
5) Went into Layer Blending and set it to just the Red Channel.
6) Masked out the center star.
7) Flattened it.
8) Continued the FFT action as if I had just finished painting.

My question is, is this actually mathematically sound, or did I get decent results simply because it would have been hard to make the photo any worse?

Thanks!
Julie

[TommyO]

Julie,
This may be a good question for bart_hickman or byRo.
They are the experts. Try sending them an note through RetouchPro.

[Kraellin]

ah, now, i'm not going to pretend to completely understand this either. from what i recall, you're looking at frequencies. FFT divides things out in a different manner from your normal channels and just happens to use the rgb channels to store its results. the red channel is the higher frequencies, if i recall. the stars are anomalies to the frequencies that FFT found. it's like crashing two frequencies into each other with a resultant 'burn'. that 'burn' doesnt belong. so, we mask it out with black. thus, when you restore the channels, the burns/anomalies are gone, masked out.

the other way to do it is when you get to the 'stars', clone them out rather than masking. you'll see it doesnt matter a twit

and as for the math, i couldnt tell you.

and, welcome to RP