On: Show HN: Resize images using the FFT

[eliteraspberrie]

Yes, the sharpness does decrease with the interpolation. I think the phenomenon is called "overshoot" and I don't know how to mitigate that.

Here is the image interpolated 4X: http://imgur.com/a/8YrxG

[kken → eliteraspberrie]

It's actually called "ringing" and can easily be explained by the fact that you convolved with a box-function in the frequency space (ideal low pass), which will transform into a sinc function in the local space.

http://www.thefouriertransform.com/pairs/box.php

[wfunction → kken]

I think you've convoluted the terminology. It's "convolve".

[kken → wfunction]

what? :)

[fat0wl → kken]

You can actually see it if you look closely -- there is a kind of striping of light/dark horizontal lines, easy to spot at the top & bottom of the image.

[dnautics → kken]

also known as the gibbs phenomenon.

[eliteraspberrie → kken]

Thanks. I'll add a window function and see if that helps.

[angersock → eliteraspberrie]

Try it with a checkerboard pattern--somewhat the pathological case.

Gray cats are somewhat smooth from a signal standpoint.

[eliteraspberrie → angersock]

Good idea. Here's the result: http://imgur.com/a/EHNH0

Edit: I'll be adding a window function soon to attenuate the ringing, it'll be interesting to see the difference that makes.

[dnautics → eliteraspberrie]

you can also try to apply a smoothening function: Usually multiplying by a gaussian in phase space will do, but this will cause your image to be blurred. What would be really impressive is if you could interpolate a pattern in the high frequency signal, and use it to "infer" values for your high frequency padding that "naturally cancel out" ringing to below the resolution of the imag.e

[angersock → eliteraspberrie]

A beautiful example of ringing. :)

[oakwhiz → eliteraspberrie]

I'm surprised that the ringing isn't that bad even without a window function.