IVTC-ing "true" hybrid NTSC material: what's the trick?

[Serg Belyansky]

OK, I'm really stumbled.

Here is short source clip (http://test.otaku.ru/eien/eien_dvd.m2v) with truly-interlaced part superimposed over telecined part (60i titles roll over 24p video, in NTSC).

For a long time I was sure that this type of material cannot be perfectly converted to progressive:


deinterlacing will give us jerky background, and
IVTC will produce jerky titles roll.

But somehow the fansubber of these series managed to do all properly: here is the short clip (http://test.otaku.ru/eien/eien_fansubbed.avi), which is 23.97, and both the background pan and credit roll are made perfectly smooth without a sign of jerkiness (aside from minor field artifacting).

(OK, titles are not identical: they a from TV and DVD version, respectively, but the nature of the sources is the same.)

My question is: what is the magic involved here?

[Anonymouses]

My question is: what is the magic involved here?

No, they just post-process deinterlaced those frames which is obvious by the fact that this a great deal of aliasing in those frames and the deinterlacing artifacts on the credits. It would also help if you could get identical sequences to match up.

[Guest]

If that were true, the credit scrolling would be jerky, and it appears smooth.

[Serg Belyansky]

Anonymouses, the sources for TV version are from several years ago and now seem to be unobtainable in the fansub world. But aside from formatting of the titles, they are identical to DVD version. I extracted the relevant parts of sources matching the background part.

[Leak]

deinterlacing will give us jerky background, and
IVTC will produce jerky titles roll.

Hmmm... the DVDs wouldn't happen to contain a credit-less opening or ending as an extra?

In that case, it's probably possible to create a difference mask to separate the scrolling credits from the video, do a "proper" framerate conversion to 24p on them and then superimpose them over the IVTCed credit-less version again. Not that I've ever tried that, but it might just work...

np: LFO - Unafraid To Linger (Sheath)

[Serg Belyansky]

Leak, the credit-less version of this final sequence is not present anywhere an the Japanese DVDs. And since the title was not released in USA, I assume that it is unobtainable, too.

Anyway, it was not available at the time of the TV version fansubbing!

[mg262]

On the mysteriously cleaned up version, does anyone else notice a period-2 alternation in the credits? Specifically, focus on a single character, single step, and see whether it seems to alternate between being slightly blurry and sharper.

[Serg Belyansky]

mg262, yes, there is alteration; but since the credits move on subpixel basis, not matching actual TV lines, I assume some period in character distortion to be natural.

[mg262]

:( ... you're right. So much for that idea. Have you tried to measure the distances the credits move every field to see if they are constant?

[Serg Belyansky]

Okay, I've fugured it out. Here's the trick, very simple and dry.

mvbob()
#check the cadence after this step; say, it is aabbb
selectevery(5, 0, 1, 2, 3) #stripping the last b
selecteven()
The background is smooth and perfect. The credit roll is OK (with minor aliasing due to deinterlacing and with high frequency jitter whick goes unnoticeable).