4:3 Letterbox to 16:9 anamorphic. Show me your methods!

[SatStorm]

The situation: Most music channels channels broadcast 4:3 while the majority of the music videos are widescreen. The same happens with most movie channels (I'm talking about Europe here).

The project: Crop the black bars from the letterboxed transmission and create from that source an anamorphic 16:9 result (mpeg 2, 720 x 576, for DVD)
The source: DVB /s channels from all around Europe. Most of them have a framesize of 544x576

As you understand, this project needs heavy filtering, resizing, etc.

How I do it:

1 - I load the 4:3 mpeg 2 DVB file to virtualdub mpeg2
2 - I add my typical filter chaine to fix what is possible from macroblocking and to delete the station logo (if it is inside the 16:9 area).
3 - I crop a total of 144 lines from the 4:3 picture. If the picture is not centered, I may have to adjust a bit the top / bottom (for example: 68 lines from top, 76 lines from the bottom)
4 - I resize the cropped / filtered result to 720 x 576 using bicubic and frameserve to TMPGenc (or whatever), so to encode to a new 16:9 anamorphic mpeg 2 for DVD.

Even if I'm pleased with the results I have with this method, I wonder what others do for this kind of convertion. I also wonder if there are other alternatives to do this. Programs dedicated to that kind of convertions for example. Technics or special filters I missing.
And I'm very concerned about the resizing here: Beyond the fact that I have to resize the horizontal lines from 544 (or whatever) to 720 (that's the easy part), I scretch vertically the 432 vertical lines to 576 again. How this can be done the best possible way? I suspect resizing ain't the best method: I need upscaling methods here!
Are there any scaling software solutions for something like this? Commercial or freeware solutions?

[guns1inger]

If it is letterboxed already, then I would do pretty much what you are doing, but do it in avisynth. I would resize with Lanczos4 (or Lamczos in vdub) rather than Bicubic, as I find it preserves details much better when resizing up. I would only use Bicubic when resizing down. But each to their own.

[SatStorm]

I use Lamczos in virtualdub, but for the opposite! When I convert 544 x 576 to 352 x 576 for example. It does a better job than bilinear on this. For resizing to 720 x 576, I use mostly biqubic because it seems better combined with msu_smart_deblocking (a neccessary filter when you encode from dvb/s interlace sources). When I convert progressive sources, lanczos is a better option.

But here, we talking about a vertical resizing of an interlace source, which I believe it is more of an upscaling. I wish to convert those 432 vertical lines to 576 vertical lines the best possible way. This method I use ain't the best, there must be a filter or something better than this.

[manono]

Hi-

But here, we talking about a vertical resizing of an interlace source

LeakKernelBob(Order=1)#if TFF, Order=0 if BFF
#***Filter Here***
Crop(0,72,0,-72)
#***Or Filter Here***
Lanczos4Resize(720,576)
#***Or Filter Here***
SeparateFields()
SelectEvery(4,0,3)#if TFF, SelectEvery(4,1,2) if BFF
Weave()
ConvertToYUY2(Interlaced=True)#for CCE, ConvertToRGB24(Interlaced=True) for TMPGEnc

Filtering interlaced sources requires special handling. Do your Cropping, Resizing, MSUDeblocking, etc. while it's bobbed, between LeakKernelBob and SeparateFields().

http://neuron2.net/kerneldeint/kerneldeint.html LeakKernelDeint at the bottom of this page

[FulciLives]

It is time to learn AviSynth my old friend

Not sure I would do it exactly the way manono did it but his way is probably better than my way anyways LOL

But I know a certain yellow fiend that doesn't care for AviSynth so ... if I recall doesn't VirtualDub(Mod) have a function similiar to separate fields? Called fields unrolled or something like that? I'm at work so ... no access to my usual video programs.

Anyways I would crop first then unroll fields then resize (on a field level not a frame level) then apply filters then roll it back again.

Or something like that LOL

- John "FulciLives" Coleman

[SatStorm]

Interesting idea here... Yes, virtualdub has the fold/unfold method to seperate the fields (it is a fuction of the de-interlace filter).

For european DVB sources, the old version of virtualdub mpeg 2 (1.5 something) handles far better everything else I tried (freeware stuff always...). Virtualdub mod is good, but not as well as virtualdub mpeg 2 for this. It is good for VOBs or SVCDs but no for direct from satellite DVB captures.

@FulciLives:The reason I can't use avisynth direct, is because the videos are widescreen but not always 16:9! They use other aspect ratios as well, so I need the fast preview that virtualdub offers to adjust stuff. Also, I have to de-logo parts of the screen, extra titles, etc...
And I'm talking for thousents of videos here...

@manono: I do all the filtering / etc before cropping and resizing (upscaling). MSU_smart deblocking must always be first on the filter chain, it doesn't work otherwise.
I'll study LeakKernelBob to see what it does and try some stuff with it. Thanks for the advice.

Other ideas?

[manono]

Hi-

@manono: I do all the filtering / etc before cropping and resizing (upscaling). MSU_smart deblocking must always be first on the filter chain, it doesn't work otherwise.
I'll study LeakKernelBob to see what it does and try some stuff with it. Thanks for the advice.

Sure, filter before cropping if you like. In my sample script I had a space set aside for filtering before the crop/resize. If you mean that you want to use it even before separating the fields, or before Smart Bobbing, you simply can't. The home page for it doesn't say anything about it being interlace aware.

If interlaced video is involved, you have to separate the fields first before doing anything. And bobbing the fields with a Smart Bobber like LeakKernelBob, filtering, and then reinterlacing is a better method than simply filtering the fields themselves and then reinterleaving.

[MovieDud]

Dvd Rebuilder can take 4:3 Letterbox and change to to 16X9. I have done it to several older dvds (non-anamorphic) and looks good. I actually prefer that than expanding the image from the HDTV. Just a thought.
MovieDud

[guns1inger]

DVD Rebuilder requires that you have an authored DVD format. It can't work with raw mpeg, .ts or DVB files.

[SatStorm]

@MovieDud: Dvd Rebuilder is for DVDs. It can be a good alternative for letterboxed movies, but not in my case. Thanks for the suggestion

@manono: Filtering the image, is not an issue for me. I use specific filters for specific sources and various combinations, with great results, tested on very demanding screens.
This vertical upscaling / resizing is what bothers me.

I changed a bit the order of the filters: MSU_Smart deblock, DeLogo, resize / crop, unfold, filtering, fold. The results are the same without unfold/fold (no suprise here, after all, the filters I use support interlace).

I'll test a bit the well known de-interlace technics combined with resizing, so to produce progressive anamorpic results from interlace letterboxed sources.

[FulciLives]

@MovieDud: Dvd Rebuilder is for DVDs. It can be a good alternative for letterboxed movies, but not in my case. Thanks for the suggestion

@manono: Filtering the image, is not an issue for me. I use specific filters for specific sources and various combinations, with great results, tested on very demanding screens.
This vertical upscaling / resizing is what bothers me.

I changed a bit the order of the filters: MSU_Smart deblock, DeLogo, resize / crop, unfold, filtering, fold. The results are the same without unfold/fold (no suprise here, after all, the filters I use support interlace).

I'll test a bit the well known de-interlace technics combined with resizing, so to produce progressive anamorpic results from interlace letterboxed sources.

You have to unfold ... resize ... then fold.

You can't resize interlaced video on the frame level ... needs to be done on the field level.

In the example above you said you resize / crop before doing the unfold which can't be good but then again I'm not sure how VirtualDub does the resizing of interlaced footage when you use it in that way.

- John "FulciLives" Coleman

[SatStorm]

Fulcilives, I have to crop first, because the croped interlace letterbox frame is what I wish to resize to anamorphic.
But you have pointed something interesting here, that needs testing. Cropping/resizing can be at seperate steps. (Us vitualdub users, can resize / crop in one step, because virtualdub do that easy)

Virtualdub has something called "null transform". I'll test filtering,crop, unfold, resize, fold and see if the result is better.

[manono]

This guy's a hard sell, eh FulciLives?

I have to crop first

You don't have to, but you can, if you're careful about how you do it it and the order of the rest of the filters.

But you have pointed something interesting here, that needs testing. Cropping/resizing can be at seperate steps.

Something made quite obvious by my earlier script. And, of course, the crop/resize can also be split up in VDub.

This vertical upscaling / resizing is what bothers me.

And it should. And that template script I gave you before is better than anything VDub filtering can produce.

Here's what some of the AviSynth big guns have to say about it:

The separatefields/weave method is better than just plain resizing but it still introduces unnecessary blurring. You have to adjust for the field shift.

(neuron2)

Using a good motion adaptive smart Bob to extract full frames from the fields and produce a double rate source stream, which you then process and finally reinterlace will give the best results for a genuine interlaced source.

IanB (AviSynth developer))
http://forum.doom9.org/showthread.php?t=113649&highlight=interlaced+resize

[FulciLives]

I'll test filtering,crop, unfold, resize, fold and see if the result is better.

That makes sense to me although I would fiter after unfold but before fold ... but trying multiple ways is the only way you will find the best way ... so you might want to "mix it up" and try a few different combinations.

As for wanting to use VirtualDub because of it's filters ... you can still use manono's AviSynth script to do the crop and resize ... but nothing else.

Then load the AviSynth script into VirtualDub. Do your VirtualDub filters as you normally would ... either saving to an uncompressed AVI (or use HuffyUV etc.) or just framserving straight to TMPGEnc Plus ... or however you normally do it.

Or do it the opposite way ... load the original file into VirtualDub and do all your filtering but no crop / resize. Save to an uncompressed AVI (or use HuffyUV etc.) then use manono's script for crop / resize and open the script directly into TMPGEnc Plus or whatever encoder you want to use.

- John "FulciLives" Coleman

P.S.

This guy's a hard sell, eh FulciLives?

Well ... you might say that ... but ... as I have learned that is just his way and I wouldn't want it any other way ... in fact I'll shit bricks if he decides to end up using AviSynth LOL

[guns1inger]

FWIW, most of the MSU filters, including Smart Deblocker, can be called from within avisynth. Many virtualdub filters can be called this way. You can easily get the paramers list by setting the filter up in virtualdub, saving the processing settings, then using notepad to cut and paste from the vcf file. This gives you the ability to adjust visually, then drop the numbers into the final script.

[manono]

Yes, and the MSU site even gives explicit instructions on how to set it up within AviSynth.

in fact I'll shit bricks if he decides to end up using AviSynth LOL

Hehe, stranger things have happened. It's good that he likes to experiment and test things out for himself. If quality is his goal, sooner or later he'll come around.

[FulciLives]

@SatStorm

What about posting a sample video ... I know it can't be very long due to file size ... but there are plenty of those free file hosting websites.

Surely you can post a 10 second long sample (maybe longer since it is afterall already compressed to MPEG-2 as it is).

Would give us something to "play" with

- John "FulciLives" Coleman

[SatStorm]

Manono, cropping / resizing in seperate steps, is not obvious for a virtualdub user like me! For years I do that in one step (that's how I learned to do it the first place) and that became more of a habbit. A habbit that I can change very easy, now that I realised it.

The reason I crop first, is because that saves me proccessing time. Crop / filter / upscale is faster in my PC than resize / filter / crop.

What I'll try, is to change to deblock/delogo/ crop / unfold / filter / upscale / fold so to see if that has better results.

You have to realise that I don't encode 2 - 3 files, but 60 - 70 mpeg 2 files in a raw. Each of them, needs seperate handling. I can't simply use the same script for all of them. But since I want more than what I have now, I'll search for the best possible alternative. If I won't succeed with virtualdub, I'll move to avisynth. But if I find a way to do that with virtualdub, I'll stick with it!

Test clips eh? I'll see what I can do with that, this weekend!

[SerbianBoss]

Manono, i think that SelectEvery(4,0,3) is for BFF not for TFF.

Because all script with 4,0,3 are ok and with 4,1,2 are jerky.

Captures are DV avi Pal interlaced.Just to mention

[SatStorm]

First tests show some improvement !
I'm excited! More test follows

[manono]

Serbian Boss-

It kind of depends on what's gone before and how you encode it after:

This returns TFF:
AssumeTFF().SeparateFields().SelectEvery(4,0,3).We ave()

This returns BFF:
AssumeTFF().SeparateFields().SelectEvery(4,1,2).We ave()

This returns BFF:
AssumeBFF().SeparateFields().SelectEvery(4,0,3).We ave()

This returns TFF:
AssumeBFF().SeparateFields().SelectEvery(4,1,2).We ave()

http://forum.doom9.org/showthread.php?p=738914&highlight=SelectEvery%284%2C0%2C3%29#post738914

I think my script above makes it TFF when done, no matter what it was originally, although if the encoding is set up correctly, the result shouldn't be jerky. In addition, he mentions something about a bug in LeakKernelDeint not found in TDeint(Mode=1), of which I was unaware. All the more reason to test out the scripts in VDub before sending to the MPEG2 encoder. Thank you for the clarification.

[SatStorm]

The things are much more complicated that I thought....
I just discovered something, which others may have notice before...

Most of the European DVB satellite channels, use something called "statistical multiplexxing". The frenchs were the firsts that did it. I thought that this has to do with the bitrate, but I was wrong... It has to do with more things.
A wierd kind of progressive / interlace switching is taking place during the broadcasting, all the time. I first noticed that years ago, but I never realised that it is used all the time.

When the source is progressive, the broadcast remains interlace but the fields are not in an offset mode between them. That makes it "progressive" without being progressive. It make it pseudo-progressive. When the source is interlace,everything remains normal.
But things can turn even more complicated than this: It seems that same channels use a threshold of some kind: Beyond a certain point, they automatically de-interlace an interlace source and make it pseudo-progressive in a way that reminds me de-interlacing by doublicate one field!

The situation is a mess: And it doesn't shows when you resize on a horizontal axis (from 544 to 720 for example), but creates HUGE problems when you resize on a vertical axis.

The only problem you have when you simply resize the horizontal axis, is that when the broadcast is pseudo progressive, the filters for the progressive sources actually work.
That's explain (to me) why my filter chains produce great results on some projects, and do nothing on others.

The only thing I can do now, is to scan each video and see if it broadcasts to interlace or pseudo-progressive. If it is pseudo-progressive, I must handle the video as progressive and feed it to the encoder that way. Progressive input / progressive output.
If it is normal interlace, I must do all those steps by seperating the fields, filter, etc, and feed it as interlace.

When the source is pseudo-progressive and I seperate fields etc, the results varies a lot... Sometimes are good, sometimes are crap. Probably I have to find the correct order to apply the filters and also when to "tick" interlace input source and when no.

I grabb direct the stream of those DVB channels, that's why I have to deal with this situation. Those capturing DVB channels in an analogue way, don't face this. It is interlace for them (probably blurred...)

Those broadcasts are even more wierd sometimes: For example: The video clip is pseudo-progressive and at the same time the inserted titles of the song are interlace!

The situation is a mess and I'm in frond of something I never imagined. Much testing must be done right now, so to find a solution.

[FulciLives]

and they say PAL is better than NTSC

[SatStorm]

Most europeans DVB channels are crap. That's another reason why those widescreen flat TVs >32" don't sell well here: If you spent 4.000 euro for a 42" HD - Ready (or true HD) screen and see how the average european DVB channel shows, it's a shock!

One thing that puzzled me the last 6 months, was why the upscaled DVDs looks so good on those screens and why even the best (on paper) DVB channels look so bad. Now, I know why: Broadcasting mess...

[SerbianBoss]

@manono

It's interesting,

I was tested long time some parametars in CCE and in script for reinterlance.

Most of my work are interlaced Pal DV avi material, bff(captured with Canopus 110). So when make script with reinterlance(with your previous help) and when using leakkernelbob i only got good result with this:

AssumeBFF().SeparateFields().SelectEvery(4,0,3).Weave()

And because i want to keep BFF in CCE i turn off output top fields and set offset to zero. Because in CCE documentation sais that if want to keep BFF you must to change offset line to zero.

So with SelectEvery(4,0,3) i got ok result on computer and on dvd players.
I will try tdeint with same parametars and say results.

[manono]

You've come a long way in a short time, eh Serbian Boss?

Because I don't capture, or use a camcoder, I rarely see BFF material. Most (but not all) interlaced DVD video is TFF.

And now I understand why the opposite setting from what I suggested works for you. I think if you did it the way I said, and then encoded for TFF, you'd get smooth playback. Instead you're doing the opposite and encoding for BFF and getting smooth playback. As long as it works, and especially as long as you understand what's going on and can adapt the scripts to your own needs, then that's the main thing. Good going.

[SatStorm]

Now I have a new question: This pseudo-progressive thing... How to output it? Progressive or interlace?

I handle the input as interlace, since the fields are present. But why to keep the output interlace? The fields are not in an offset between them!

Your opinion?

[manono]

Hi-

Your opinion?

My opinion is that maybe you should upload a small unprocessed section of 10-15 seconds so we can have a look at what you're trying to describe. Open your source in DGIndex, use the "[" and "]" buttons to isolate a section with movement/motion, go File->Save Project and Demux Video, and upload the resulting M2V here:

http://rapidshare.de/

When done uploading, it'll give you a link which you then post in this thread.

[SatStorm]

Ok, here is a sample: http://dl6.megauploads.org/download.php?id=65E9FFDB

It comes from "Rapture TV", a UK based channel. It is a direct DVB rip from the satellite.
It is "Pseudo" progressive, 16:9 letterboxed. The picture quality is crap as usual...


And here is how it ends up using virtualdub filters, etc:
http://dl1.megauploads.org/download.php?id=796D5F30

Not much of improvement, and it looks blured too!

[manono]

Hi-

Personally, I think you did about as good a job as could be expected with that really lousy source. You did as well as can be expected on the blocks. You didn't overdo it with the deblocker, as there are still some blocks to be seen in some places. Part of the blurriness is from the deblocking and part from the conversion from 4:3 to 16:9 (and part from other filters?). At the most I might suggest a slight sharpening at the very end, but I don't think it will help much.

It's not "pseudo" progressive. It's progressive. I'm not real sure why you think there's anything unusual about it. Maybe it really was shot on video and deinterlaced (although I doubt it), but it's true progressive now. Sure, it was encoded as interlaced, but the source is progressive and can be treated as such. Where you see the frames repeating, I think that's just an effect used to slow down the girls' movement. I guess that discussion earlier in the thread about filtering interlaced sources was for something else.

Oh, and the source isn't 16:9 letterboxed, but 4:3 letterboxed, or, as I prefer to call it, widescreen 4:3.

If it is normal interlace, I must do all those steps by seperating the fields, filter, etc, and feed it as interlace.

If you have any samples of those, I wouldn't mind seeing. Thanks for the sample. Those broadcasts sure make your life difficult, don't they?