How to convert double blurred IVTC?

[odyssey]

Seeking this forum leads to numorous NTSC-to-PAL threads and some involving blurred telecine, but nothing really covers the issue with double-blurring and recovering the hidden picture.

I wanted to convert a clip from NTSC 29.97 (appearently Progressive) to PAL, and ran into this problem, and a thread stating that this was impossible. So I thought I'd do a little research myself!

The image sequence are as follows: A B BC CD D E EF FG G ... hence C and G pictures here are hidden in each other. See the example below:

http://img474.imageshack.us/img474/5751/plutoblurred0is.gif (http://imageshack.us)

I'm not an expert at deinterlacing, but a little experimenting with Photoshop shows that taking odd lines from the first picture, and even lines from the second picture, reveals the original picture.

Is this really impossible with filters??

[Joel Cairo]

Well, if that blend pattern stayed constant, and you had no dropped frames in the capture, you should just be able to SeparateFields, and do **this**, I think:

SelectEvery(8,0,1,2,3,5,6)

And then weave them accordingly... this should pick out the proper 3 progressive frames out of every four, and discard the unnecessarily duplicated fields.

This, of course, is entirely limited to the question of recovering the progressive frames, and is irrespective of the final framerate of your sequence. There's probably a more elegant solution, but this is one that immediately comes to mind.

-Kevin

[odyssey]

I think I get the point, but wouldn't it disturb the intact frames, to be SeperateFields()'ed when it's not an interlaced source?

I looked a little on the ComplimentParity() filters among other, but they seem to process the entire stream, and not just individual frames. Am I missing something? (Still quite new to AviSynth scripting)...

Are there anyway you can experiment these deinterlacing settings on-the-fly f.ex. in VirtualDub?

[Mug Funky]

i don't get your problem. is there some reason telecide or TIVTC doesn't work?

or even set "force film" when you make the d2v file.

[odyssey]

i don't get your problem. is there some reason telecide or TIVTC doesn't work?

or even set "force film" when you make the d2v file.

Hummmmm guess not then ???? :thanks:

[Mug Funky]

no problem. i was confused because the thread title said "double blurred" but the pics you posted showed perfectly intact 3:2 pulldown (that is 3 progressive frames, 2 interlaced, 3 progressive, 2 interlaced... or more specifically 3:2 is for the pattern of fields per original film frame, so you get 3 fields of 1 frame, then 2 fields from the next, etc.).

if you had actually blurred fields, then most likely you're s*** out of luck (this happens when you vertically resize an interlaced clip without due care and attention).

[edit]

btw, it's possible to construct an automatic inverse-telecine function in avisynth using only internal functions (no plugins), but it's veeery slow this way. i tried it once for fun. all you need is to compare a field-order swapped frame with the original, and keep the least combed frame. then drop the frame in every cycle of 5 that is the same as the previous frame.

[odyssey]

no problem. i was confused because the thread title said "double blurred" but the pics you posted showed perfectly intact 3:2 pulldown (that is 3 progressive frames, 2 interlaced, 3 progressive, 2 interlaced... or more specifically 3:2 is for the pattern of fields per original film frame, so you get 3 fields of 1 frame, then 2 fields from the next, etc.).

Now that makes sense :D I wish interlacing was killed a long time ago...

[Mug Funky]

well, it's okay if your destination format allows interlace...

the transition is painful as hell though - in the analog editing days, field order wasn't an issue - it all just worked. then digital comes in to pro video world, and it's a little more complicated, but still it all was consistent and all worked.

then prosumer/consumer/etc stuff came in. NLE programs defaulted to the wrong field order, standards were mixed willy-nilly, and the software didn't provide adequate "idiot-proofing". combining that with lossy compression being involved, and people editing without using an actual television, and suddenly everybody hates interlace because it always seems to come out wrong. but there's very few inherent problems with it - line flicker and visible combing on certain types of motion are really the main visual drawbacks. it's just more difficult to work with.

but that said, if you're converting pure film to PAL, definitely IVTC using decomb or tivtc or whatever, then speed up to 25fps. it's the best way, and it'll encode faster too :)

btw, welcome to the big geeky world of avisynth :)