Perfect deinterlacing of PAL DVD video?

[mantis2k]

What's the most perfect up-to-date method of de-interlacing PAL DVD video? Is there any guides available for newbies like myself? I heard someone mention Avisynth and Bob? Is that really the best method out there, or is there a more perfect way? I've never even used AviSynth before, so any help would be much appreciated!

Also, is it best to deinterlace the MPEG video @ source resolution or target resolution?

[Zyphon]

This is an identical post to the one that you posted in the link below:

You shouldn't really double post the same question.

The anser was given there. :)

http://forum.doom9.org/showthread.php?t=97677

[Guest]

@mantis2k

Please read and follow forum rules, specifically rule 8: no cross-posting. Thank you.
:readrule:

[Mug Funky]

"bob" more describes a double-framerate deinterlace than a filter. there is a bob filter internal to avisynth, but it is very basic, and is really not all that useful if perfection is what you're after (but it can of course be used as a building block for a more sophisticated deinterlace).

then there's "smart bob". this usually describes a motion-adaptive deinterlacer - that is, if an area of the screen is static, then there's no need to deinterlace it, and so it is left as-is while the moving parts are deinterlaced. this greatly improves vertical resolution and stops the screen from shimmering at playback on a progressive device (ie computer monitor, or a HDTV). as for the moving parts, there's several ways to try to interpolate in the most detail-preserving way possible.

the closest deinterlace to perfection is a motion-compensated approach (not to be confused with motion-adaptive. these are different things). this means to effectively to fill the missing lines by moving the other field on top of the current one, and interpolating the (very small) differences left behind. typically these methods are artefact-prone, slow, and/or incredibly expensive. since avisynth is free, all you get is artefact-prone and slow solutions :). but the right settings on the right source can be brilliant - really not much different from a bona-fide progressive source.

deinterlacers to look up:

- Kerneldeint (and Kernelbob), or the speed optimised LeakKernelBob which is almost as fast as regular bob()
- Tdeint - slower than the above, but generally better result, and brilliant on sources with mixed-in film content.
- mvBob - an attempt at a motion-compensated deinterlacer, with built-in artefact protection (so you wont get much weirdness when it doesn't behave itself).

i can't say which is best, because they all have their strengths (and the forum rules don't allow it). leakkernelbob is great for standards-conversion - it's hella fast and gives a result very similar to standalone hardware converters. Tdeint is good when you've got a crap source and the field-order is b0rked (it happens more than you might think). it's less prone to artefacting when fed the wrong field-order, so i use it when leakkernelbob fails.

MVbob isnt much good for anything at the moment unless you've got a cluster of Cell processors, or are extremely patient and don't mind waiting days for an encode. really nice results though. it's probably worth using on short films if you need to squeeze every pixel's worth of information.

forum search is your friend... :)

[mg262]

the speed optimised LeakKernelBob ... is almost as fast as regular bob()What setup are you using? Just last night I found out the hard way that it was substantially faster in the context I tried it in;* that is using no settings (except order) on a 2.4GHz P4, and comparing with Bob from AVISynth v2.55. I haven't tried a direct comparison though...

*I think you can guess what that is...

[Mug Funky]

hehe, well, i say it as a sweeping generalisation. it's probably a fair bit slower than bob, but on some content it can even be faster (ie talking heads with a high-ish threshold).

i usually use it at a low threshold (like 1-2) for standards-conversion - occasional very mild shimmer is more acceptable than weird leftover combing being resized. at low thresholds it's slower because it's doing more deinterlacing.