sinc

The technical answer is that you don't interpolate the samples with lines, but with the sinc function.

The approach, bandlimited interpolation, is executed in the frequency domain here and in the space domain in sinc interpolation.

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.

You have to choose Sinc Lancoz 3 because that is the interpolation option that will preserve the data in the jpeg pixels.

The article illustrates antialiasing, but chooses to use sinc-based circular windows, not square windows.

ImageMagick can do this too: convert -filter Sinc -resize 200% input.png output.png However, as pointed out by other commenters, sinc by itself isn't the greatest support function for enlargement due to the ringing artifacts [0].

The sinc function is sinusoidal and so it more naturally approximates waves.

The frequency response of a box filter is the sinc function, which has a lot of energy above the Nyquist of this sampling rate.

The ideal sinc waveform [..] would require infinite delay.

However, in practical use, we immediately have to compromise by at least windowing the sinc; thus we get into various approximations to minimize artifacts.

Sample depth roughly tells you how far away a sample can be before the magnitude of the ideal sinc function used to reconstruct the continuous signal falls below your quantization threshold.

I get that Eich donated to support Prop 8 in 2008; it caught my attentions traight away sinc I was on the opposite sid eof that debate.

It's also "impossible" to upscale an image, but the difference between bilinear and bicubic and sinc upscaling is readily apparent.

The smooth fall-off of CRT's pixel "point spread function" actually resembles a bit more the ideal sinc filter for reconstruction of bandlimtied signals.

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.

The reason linear filters don't work for upscaling, say a Lanczos or Sinc, is that the base supposition made for generating the filters is that the source is "bandlimited" -- that is, all images are samples of some source which can at most vary once every pixel.

Sinc interpolation is usually also windowed with a dampening factor on the tails of your filter.