Hacker News Reader

Comment by andrewcooke | Hacker News Reader

is this the same as sinc-interpolation? (effectively; that happens in "real" space, while this seems to be in frequency space)

View on HN

As far as I can tell from the code, this is exactly the same. The approach, bandlimited interpolation, is executed in the frequency domain here and in the space domain in sinc interpolation.

Bandlimited interpolation is "correct" in some technical sense for sampled signals and images, making some assumptions about the sampling procedure. That doesn't imply it's perceptually optimal, though. If your goal is "best approximation of what I sampled given the available data", then bandlimited interpolation is good. If your goal is "image that looks good", it probably isn't.

I'm not crapping on this contribution, though, it's always cool to see more signals stuff open sourced.

To be more clear: a sinc reconstruction is correct in the sense that if the original signal was bandlimited before being sampled, the sinc reconstruction will perfectly reproduce the bandlimited input signal. All other reconstructions can produce aliasing (how much and whether it is objectionable obviously depends on the input signal). In cases where the original signal was not bandlimited concepts like "perceptually optimal" come into play and a different reconstruction filter may be preferable.

Sinc interpolation is usually also windowed with a dampening factor on the tails of your filter. I think the ringing on the left edge of the image shows the lack of windowing.

Yes, I think so, although it's been a couple years since I recall the proof.

Reply on news.ycombinator.com