Using NTSC (16:9) and PAL (4:3) sources to restore a film...

[takearushfan]

Hey all,
Despite the initial difficulty, I managed to do a good enough job to show you (in the attached sample) what I'm trying to do.
The region 1 DVD of The Vault Of Horror is edited, but looks and sounds great. The region 2 version is unedited, but looks and sounds awful. To further complicate things, the R1 version is widescreen, the R2 version fullscreen.
Anyway, you can see from the attached file that I'm trying to piece between the two versions, for the sake of restoring the material that was cut out. The difficulty comes from the AR issue.
Is there a way to "even out" the ratios, so you can't tell (as easily) that I'm jumping between full/widescreen?
As for the picture/audio quality of the R2 portions, I'll deal with that later. For now I'm just wondering how to properly deal with the "stretched"/"squished" AR problem.
Thanks SO much for your help
- Justin

[jagabo]

You don't want to pillarbox the 4:3 sections? The panning shot near the end is really jerky. Every 6th frame is a duplicate.

[manono]

Is there a way to "even out" the ratios, so you can't tell (as easily) that I'm jumping between full/widescreen?

There's some evidence to support a claim that the 1.33:1 source was matted for 1.85:1 display in the movie theater. So, if that R2 4:3 release used the source 1.33.1 print (rather than cropping down a 1.85.1 version), you could crop off the top and bottoms and encode for 16:9 to match the R1 DVD:

http://www.imdb.com/title/tt0070868/technical

And do a proper PAL2NTSC conversion. Duping frames and removing unique frames isn't the way to do it.

The aspect ratio is 1.53:1 (696x456)? Really? What'd you do? Crop from the sides of the NTSC version and from the top and bottom of the PAL version to make the AR for both the same?

[intracube]

The aspect ratio is 1.53:1 (696x456)? Really? What'd you do? Crop from the sides of the NTSC version and from the top and bottom of the PAL version to make the AR for both the same?

The NTSC/PAL segments are distorted. People are too thin in the NTSC segments and fat in the PAL bits. Looks like takearushfan just distorted the two clips to fit the frame.

It's possible the 4:3 release is neither the full image from the 1.33 master, nor the 1.85 version with the sides missing - it could be somewhere between the two...

[jagabo]

Maybe he was shooting for ~half way between the two?

[intracube]

Maybe he was shooting for ~half way between the two?

Distorting the proportions of the whole film for a few minutes of inserts... wouldn't be my choice

The 4:3 version is on YouTube:
http://www.youtube.com/watch?v=ZbLIUwXgJ18#t=9m37

Framing comparisons:


The position of both frames looks fairly consistent in all the shots I've seen.

[takearushfan]

This is why I need to check my inbox more often

No, the original resolution wasn't 696x456. Your assumptions are correct. Though the sources were the typical native ratios (R1 - 720x480/R2 - 720x576), I indeed cropped both to fit the same frame size.

I don't know how to get the R1 16:9 and R2 4:3 sources as close as possible. I agree that cropping isn't ideal but I don't know... heck, I don't know what I'm doing in general :/

The attached clips are completely unaltered, if you want to see the differences.

[takearushfan]

You don't want to pillarbox the 4:3 sections? The panning shot near the end is really jerky. Every 6th frame is a duplicate.

I've never used that technique; not sure how to.

[takearushfan]

"And do a proper PAL2NTSC conversion. Duping frames and removing unique frames isn't the way to do it."

I didn't know I even did that. How did I manage to dupe some and remove some of the unique ones?

[takearushfan]

All I did was use the same script for both VOBs: MPEG2Source, TFM, TDecimate. The difference is that I added LanczosResize(720, 480) to the PAL script. Then, as you noticed, I cropped both sources to fit the same frame.

I'm sure that I did one to a million things wrong. That's why I turned here for help

[jagabo]

You should have posted the original MPEG 2 clips, not Xvid AVI (especially not interlaced sources encoded progressive!). And a long panning shot would be better.

TFM().TDecimate() is the right thing to do with the NTSC clip to restore the original 23.976 fps film frames. The PAL clip should have gone through Something like TFM().AssumeFPS(24000,1001).BilinearResize(540,480 ).AddBorders(90,0,90,0). But that would require that you change the audio length of the PAL clips.

ntsc=AviSource("NTSC_CLIP.avi").TFM().TDecimate()
pal=AviSource("PAL_CLIP.avi").TFM().AssumeFPS(24000,1001).BilinearResize(540,480).AddBorders(90,0,90,0)
return(ntsc+pal)

Instead of pillarboxing the 4:3 PAL you could pan-and-scan a 16:9 portion:

ntsc=AviSource("NTSC_CLIP.avi").TFM().TDecimate()
pal=AviSource("PAL_CLIP.avi").TFM().AssumeFPS(24000,1001).BilinearResize(720,480,0,108,720,432)
return(ntsc+pal)

That's not quite perfect because the cropped NTSC version has a slightly wider portion of the frame than is available in the PAL version (as you can see from intracube's post).

Another approach would be to leave the PAL portions at 25 fps, encode them as MPEG 2, then apply 3:2:3:2:2 pulldown flags with DgPulldown. Encode all the NTSC clips with 3:2 pulldown. Then edit them all together with a non-reencodnig editor. That way the audio of the PAL clips won't have to be changed. This will lead to noticeable audio pitch changes though (assuming the usual PAL speedup technique was used).

Obviously, the levels and colors need work too.

[intracube]

That's not quite perfect because the cropped NTSC version has a slightly wider portion of the frame than is available in the PAL version (as you can see from intracube's post).

I'm not sure which NTSC clip you're comparing (EXAMPLE.avi or NTSC_CLIP.avi). Comparing NTSC_CLIP.avi and PAL_CLIP.avi to EXAMPLE.avi, the latter has the black bars cropped off the PAL sections (702 vs 720 issue) and therefore cropped some useful detail from the sides of the 1.85:1 version (the original is full width 720px). Also a very small amount has been cropped from the top of the NTSC segments. So my examples in post #6 aren't completely accurate. Hope that makes sense.

Updated comparison (only minor differences):

*1.85 area represents 720x463 area of the 720x480 frame (NTSC_CLIP.avi)
*4:3 area represents 702x576 area of the 720x576 frame (PAL_CLIP.avi)

Three ways the PAL segments could be formatted to fit the NTSC film:
1. Pillarbox 4:3 segments within the 1.85 (or 16:9 frame - hardly makes a difference)
(you won't loose any of the picture, but you'll see more at the top/bottom than was ever in the 1.85 version)
2. Partial crop. Crop the 4:3 segments to show the same vertically as the 1.85 version.
(this will leave small black bars at the sides)
3. Full crop. Expand the 4:3 version to fill the width of the 1.85 frame and centre vertically as best as possible to match.
(framing will be tighter/zoomed in)

[manono]

All I did was use the same script for both VOBs: MPEG2Source, TFM, TDecimate.

You removed unique frames on the PAL clip with the TFM.TDecimate by bringing it down to 20fps. You added all the dupe frames when it was brought back up to 23.976fps (in the encoder, maybe). Together they made it play especially jerky.

You either resize and encode for 25fps followed by a 25->29.97fps pulldown, as jagabo described. Or you resize and slow both the video and the audio to film speed, encode as progressive 23.976 followed by regular 3:2 pulldown, as he also described. There are no other (acceptable) ways to do this PAL2NTSC conversion, in my opinion.

[takearushfan]

"You should have posted the original MPEG 2 clips, not Xvid AVI (especially not interlaced sources encoded progressive!). And a long panning shot would be better."

Here they are. The panned shot is included.

[takearushfan]

I also forgot to mention (when I said "all I did was") that I made sure both the NTSC and PAL clips were exactly 23.976000 ... VDMod was being picky, as the one of the streams was (I believe) 23.976020 :/

[takearushfan]

Hi again...

I figure the best thing to do (to benefit us all) is explain exactly what I'm aiming for and also be completely honest, so here goes:

What I'm trying to do -

End up with the full film remaining as "NTSCish" as possible, while using the PAL shots where necessary. Besides the video issues we've discussed, there's also the problem with the audio. See, I'll need to use the PAL's audio at times, as the NTSC version obviously cuts out the R-rated audio during the corresponding video cuts. However, the PAL audio is sped up a bit compared to the NTSC; not sure what to do about that. As for the color issue(s), I was considering using the "Colourlike" filter; might that be of good use?

There's the "what I'm aiming for" part. Here's the "completely honest" part -

Though I'm no expert, I'm much more of an audio person than a video person. I'm embarrassed to admit that I know close to nothing about video. Therefore, when I read terms like "pillarbox", "3:2 pulldown", etc., I'm pretty much lost. You guys wouldn't have replied to my initial post in the first place if you weren't considerate enough to lend a helping hand to a "lost soul", so I hope my admitting ignorance doesn't turn you off.

Okay... that said:

How would you recommend approaching this project, in layman terms?

Thanks MUCH again

[jagabo]

The NTSC version is encoded progressive with 3:2 pulldown flags. The PAL version is encoded interlaced but has the top and bottom fields of the original film frames in adjacent video frames. The basic scripts I gave earlier work with these clips too. Except you have to use Mpeg2Source() instead of AviSource().

ntsc=Mpeg2Source("NTSC_VTS_01_1.d2v").TFM(d2v="NTSC_VTS_01_1.d2v").TDecimate() 
pal=Mpeg2Source("PAL_VTS_01_1.d2v").TFM().BilinearResize(720,480,0,108,720,432).AssumeFPS(24000,1001)
return(ntsc+pal)

I built the d2v index file for the NTSC clip with pulldown enabled. If the entire DVD is progressive you can create the index in forced film mode and skip the TFM().TDemimate() steps. If the entire PAL video is out of phase frames you can use SeparateFields().Trim(1,0).Weave() instead of TFM() to reconstruct the film frames. (TFM() occasionally makes mistakes so the alternate method is preferable if the entire movie is out of phase.)

Xvid AVI of above script attached.

[takearushfan]

Hm...

And what about keeping the wave audio not only in sync, but at the same pitch?

[jagabo]

And what about keeping the wave audio not only in sync, but at the same pitch?

Since AssumeFPS(24000,1001) is reducing the speed of the PAL clips you have to stretch the audio by the same amount (25*1001/24000). That will also restore the pitch.

[intracube]

takearushfan; How do you want the 4:3 PAL sections to be formatted? Take a look at the second image in post #12 with the three framing options.

Another thing to watch out for; even if you slow down the PAL version to match the pitch of the NTSC video, the volume is likely to be at a different level. If I were doing this, I'd pre-convert the PAL clips to NTSC and reformat to 16:9/1.85:1 (using one of the methods suggested in this thread, or something similar), then import them into a video editor - which will make it easy to crossfade the audio tracks seamlessly and correct the worst of the colour differences.

[takearushfan]

takearushfan; How do you want the 4:3 PAL sections to be formatted? Take a look at the second image in post #12 with the three framing options.

Another thing to watch out for; even if you slow down the PAL version to match the pitch of the NTSC video, the volume is likely to be at a different level. If I were doing this, I'd pre-convert the PAL clips to NTSC and reformat to 16:9/1.85:1 (using one of the methods suggested in this thread, or something similar), then import them into a video editor - which will make it easy to crossfade the audio tracks seamlessly and correct the worst of the colour differences.

I think I like option #2, the partial crop.
I mentioned trying Colourlike but noone seemed to chime in on their opinion of it. Good idea, bad, somewhere in between?

[poisondeathray]

I mentioned trying Colourlike but noone seemed to chime in on their opinion of it. Good idea, bad, somewhere in between?

My opinion: Somewhere in between. It gets you roughly in the ball park, but you still have to fine tune stuff after

[takearushfan]

I mentioned trying Colourlike but noone seemed to chime in on their opinion of it. Good idea, bad, somewhere in between?

My opinion: Somewhere in between. It gets you roughly in the ball park, but you still have to fine tune stuff after

I'm surprised that Colourlike worked as well as it did, albeit all I attached is a single frame comparison. When I play the actual clip, it does have some, I don't even know what to call it... it becomes a bit grainy/pixelated in some places.

[intracube]

I mentioned trying Colourlike but noone seemed to chime in on their opinion of it. Good idea, bad, somewhere in between?

I haven't used Colourlike (or Avisynth). Standard Photoshop curves can mostly normalize the colours. Your experiments with Colourlike look good - but there's a bit too much blue in the shadows.

I've experimented with several ways to format the video to look like option 2. There were a few inconsistencies between the two versions, so I've had to tweak the numbers slightly, but I won't bore you with the details.

The following operations should format the PAL clips to look similar to example 2:
- crop to 700x448 (offset from top left - 12px across, 99px down)
- scale 649x463
- pad 36px left, 6px top, 35px right, 11px bottom

I'm sure jagabo/poisondeathray can suggest the right combination of Avisynth filters to use for this.

[takearushfan]

I mentioned trying Colourlike but noone seemed to chime in on their opinion of it. Good idea, bad, somewhere in between?

I haven't used Colourlike (or Avisynth). Standard Photoshop curves can mostly normalize the colours. Your experiments with Colourlike look good - but there's a bit too much blue in the shadows.

I've experimented with several ways to format the video to look like option 2. There were a few inconsistencies between the two versions, so I've had to tweak the numbers slightly, but I won't bore you with the details.

The following operations should format the PAL clips to look similar to example 2:
- crop to 700x448 (offset from top left - 12px across, 99px down)
- scale 649x463
- pad 36px left, 6px top, 35px right, 11px bottom

I'm sure jagabo/poisondeathray can suggest the right combination of Avisynth filters to use for this.

Thanks. I'll try it out when I get out of my groggy state.
As for jagabo/poisondeathray, the crucial part is if they're still interested enough to help

[takearushfan]

"I've experimented with several ways to format the video to look like option 2. There were a few inconsistencies between the two versions, so I've had to tweak the numbers slightly, but I won't bore you with the details.

The following operations should format the PAL clips to look similar to example 2:
- crop to 700x448 (offset from top left - 12px across, 99px down)
- scale 649x463
- pad 36px left, 6px top, 35px right, 11px bottom"

Um, if the PAL source was originally 720x576 then you've already confused me at step #1. I'm assuming it has to do with the details you don't want to bore me with, heh. Would you recommend doing the crop with Avisynth or directly in VDub?

[manono]

Even if permitted to (which you won't if keeping to the original YV12 colorspace), you don't want to crop and/or resize using odd numbers.

[takearushfan]

Even if permitted to (which you won't if keeping to the original YV12 colorspace), you don't want to crop and/or resize using odd numbers.

Yep, I've noticed that unless using a lossless codec, it doesn't "agree" with odd numbers. I'm using lossless for what I suppose you could call "pre-production", but it'll have to be eventually be lossy in "post".

[manono]

Well, I use lossless myself all the time, from DVD sources to YV12 or YUY2 Lagarith AVI, depending on the eventual output, so the crop rules still apply. I don't know why you want to convert to RGB, but maybe there's a reason:

http://avisynth.org/mediawiki/Crop

You're starting with DVD and intend to make another DVD, right? So, unless you have to use some VDub RGB filter (Colorlike?), the RGB conversion shouldn't be necessary. Anyway, when converting back to YV12 for DVD, any odd number crops and/or resizes will screw things up.

[takearushfan]

Well, I use lossless myself all the time, from DVD sources to YV12 or YUY2 Lagarith AVI, depending on the eventual output, so the crop rules still apply. I don't know why you want to convert to RGB, but maybe there's a reason:

http://avisynth.org/mediawiki/Crop

You're starting with DVD and intend to make another DVD, right? So, unless you have to use some VDub RGB filter (Colorlike?), the RGB conversion shouldn't be necessary. Anyway, when converting back to YV12 for DVD, any odd number crops and/or resizes will screw things up.

I don't want to do anything other than what will provide the results I'm looking for. If this can be done without converting to RGB then no, I don't have a reason to do it.

I'd like to create a DVD from it, yes. If it ends up being too much of a pain/too complex then I'll settle for a lossless AVI, despite the size.