Lanczos Resizing

[mp34ever]

At the moment avisynth has lanczos3 and 4 resizing and other programs have lanczos5 resizing so i was thinking maybe there could be some development to make a fast lanczos6 maybe 5 12tap filter resizer wich is fast?

[IanB]

mp34ever,

The current resizer core could be taught to do Lanczos5/6 interpolation but I am not sure truncated sinc(x) windowing is the best solution for a this order of filter. The compromise of Lanczos coefficients by choice tend to ring as to favour sharpness.

This is not always entirely satisfactory and the higher the order the more ripples you will get. With a higher order you should be making a better compromise of sharpness versus ringing than this. Any budding mathemagicians wish to bless our coefficients.

The current core generates close to optimal mmx code for every resizer but the higher the order the more operations per pixel required and the exponentially slower it will be. Also above a certian complexity you risk running out of cache lines, but I think a 6th order filter would be okay.

IanB