1950's Regular-8mm film to digital HD

[videoFred]

This is the deal: BigDog660 has a nice collection of regular-8mm family films, shot in the 1950's. And I am going to transfer his footage to digital. The quality of this footage is not to bad. But colors begin to fade out, and some scenes are under or over exposed. The film grain is pretty heavy, typical for the used stock in that period, and a regular-8 frame is very tiny of cource.

We are starting with the maximum resolution of my transfer camera: 1024 x 768. The original AVI files are in uncompressed RGB24 format, progressive, 8 bits, frame -by- frame. This means: one film frame = one AVI frame. Bigdog want the best quality possible and his end target is HD.

But let me first show a picture(downscaled for the Forum):
I have used my film restoring script of cource.

http://users.telenet.be/ho-slotcars/DOOM/ron_comp_003.jpg

And now for the questions:

- Keeping in mind my originals are 1024 x 768 already, what HD format should I go for?
- Would it not be better to let the player or TV do the upscaling?

- Play speed of R8 film is between 16 and 18fps. The original AVI files are 18fps (or whatever) progressive.
What do you all suggest for the frame rate convertion?

- Is there any difference beween US HD and Europe HD? (Bigdog is in the USA)

All HD information in general is very welcome, I do not have any experience with HD and/or upscaling.


Thank you all in advance, ;)
Fred.

[2Bdecided]

First, an obvious suggestion: for an archive of his films, provide something as close to the source as possible - either the "original" or the (wonderful!) AVIsynth script restored version of it. No upscaling, frame rate conversion etc.

Then, to provide a version to watch in HD, you can either downscale a little to 960x720 and pillarbox to 1280x720, or upscale to 1440x1080 and pillarbox to 1920x1080. (Assuming 4x3 source - is that what you have?). If you think that real resolution will be lost going down to 720p, go up to 1080 - but I suspect 720p is fine.

For 720p, I'd use simple frame duplication to get it up to 60p (actually 59.94). A decent encoder isn't going to waste bits encoding the duplicate frames, and it makes it BluRay compliant.

For 1080, you can either use an irregular pulldown pattern to go to 60i, or frame duplication to go to 30p (or even 24p). I think I'd just duplicate every frame and call it 30p or duplicate every second frame and call it 24p - pros and cons with either: 15p>30p = slower running and needs 60i encode for BluRay; 16p>24p = more stutter.


You've also demonstrated some nice frame rate conversion methods - 18fps looks very stuttery to the modern viewer - but whether you want to inflict this (sometimes) artefact-prone process on this footage is another matter.


My suggestion is 720p60. It might be the only version you need to provide - i.e. close enough to "archive" to negate the need for the archive version I suggested at the top!

Cheers,
David.

[smok3]

why 60p?

> 18*3
ans = 54
> 16*3
ans = 48

i would triplicate to get close to 50p in both cases (and call that assume 50fps). edit2: i wouldn't know how to predict what will this do to a built in LCD 'superflow' engines...
p.s. - If you later decide to do a standard DVD for example, then you could make 50i version easily.

edit; i would crop some and pillarbox some, so there is no need to resize;
AddBorders(128,0,128,0) # pad width 1280
crop (0,24,0,-24) # crop height 720 - you loose 48 pixels

[videoFred]

Hi David,


First, an obvious suggestion: for an archive of his films, provide something as close to the source as possible

Yes of cource. We will keep the (huge) originals as they are on portable HD.



Then, to provide a version to watch in HD, you can either downscale a little to 960x720 and pillarbox to 1280x720,

This is a very interesting option because I have to crop the stabilized clip anyhow. And pillarboxing is imho the best option to go from 4:3 to 16:9.



For 720p, I'd use simple frame duplication to get it up to 60p (actually 59.94). A decent encoder isn't going to waste bits encoding the duplicate frames, and it makes it BluRay compliant.

Wait a moment.... I can not set play speed (assumefps() )to 15, then changefps() to 60, can I? This would show the same frame 4 times!


You've also demonstrated some nice frame rate conversion methods - 18fps looks very stuttery to the modern viewer - but whether you want to inflict this (sometimes) artefact-prone process on this footage is another matter.

This means interpolation: MVFlowFps() with possible artefacts indeed.


Thank you for the input David, :)
Fred.

[videoFred]

Hello Smok3,


i would triplicate to get close to 50p in both cases (and call that assume 50fps). edit2: i wouldn't know how to predict what will this do to a built in LCD 'superflow' engines...

I could use assumefps(16.6666) followed by changefps(50)... But that will look awful on computer...


edit; i would crop some and pillarbox some, so there is no need to resize;
AddBorders(128,0,128,0) # pad width 1280
crop (0,24,0,-24) # crop height 720 - you loose 48 pixels

That's a very good hint but I must crop my source first because of the stabilising....

Fred.

[smok3]

a. right, so basically i assume you are resizing anyway?
b. I guess the proper deshaker can be forced to greater output resolution?
c. could you do a deshaking/stabilizing and frame recomposing at the same step? - in that case i would just fill the entire 1280x720 - assuming the amateurish composing of the original that is, example;

http://blog.somestuff.org/images/recompose_to_16-9.jpg

[videoFred]

a. right, so basically i assume you are resizing anyway?

Only if I have to!


b. I guess the proper deshaker can be forced to greater output resolution?

I'm using Depan().. It has a 'zoom' option but I have never used it. It's not resizing but at least there will be no need to crop afterwards (if it works).


in that case i would just fill the entire 1280x720 - assuming the amateurish composing of the original that is, example:

That looks pretty good to me. :)
Strange how the aspect ratio makes it look like it was made in 2009!

Fred.

[2Bdecided]

I think cropping 4x3 material to 16x9 is a really bad idea. You lose content, and in this case you lose resolution (and you have none to spare!). You'd also have to do it manually - just because the interesting part is that the bottom of this frame doesn't mean it stays there - in the next shot the same cropping will probably remove everyone's head.

Note that cropping the source to get 16x9 is completely out-of-favour on BluRay - they're releasing lots of content strictly in the original aspect ratio, including 4x3 (1.33:1), 1.37:1, 1.66:1, and 1.85:1. None of these formats is 16x9, so they pillar box it. This is the opposite to the attitude on many broadcasts and some DVDs which try to avoid black bars at all costs.

Why 60p? Because the user is in America, so 50i is no use to them. 54, 48, etc etc are not BluRau compatible, and it would be silly to make an HD version that wasn't ready to drop onto BluRay.

For 60p, I would just do assumefps(16) (or whatever - it doesn't have to result in a simple pattern - go for the correct speed if 15fps is visibly slow) changefps("ntsc_double").

Yes, 4 duplicate frames. It doesn't matter at all if you're going to encode x264 (which I'm guessing you, or someone, will).

Hope this helps.

Cheers,
David.

[2Bdecided]

btw, isn't that frame quite black crushed? Maybe it's my monitor! There's a lot of shadow detail that I can only see my increasing the gamma.

Cheers,
David.

[videoFred]

The blacks are pretty OK, but I have set the gamma a bit to low. :scared:
But it looks fine on my Dell monitor....

Fred.

[smok3]

Ok, then there is reason for 60p (i missed the us part somehow...)

Any professionaly done production should stay in the original aspect, i agree (seems to be not the case here).

Depan or any other stabilizer must resize (zoom in, out, rotate, etc), so if you output to something big enough to later manually shot-by-shot fill the 1280x720, then i would pick that in this case.

[2Bdecided]

On UK TV (e.g. the BBC) we rarely get 4x3 as-is these days when there's been an opportunity to mess it up. Entire 4x3 programmes are usually repeated as-is, but whenever the footage is re-used or edited, even if there's a compilation consisting entirely of 4x3 content, it's almost always presented in 14x9 or 16x9 SD.

Sometimes it looks OK (though the picture looks undesirably soft), but sometimes it looks horrible. Then again, in the case of professional productions, the original framing was usually "good" (not so most amateur footage), and I'm not a typical person when it comes to watching video (picky picky picky!).

If you're going to crop it to 16x9 scene by scene, what do you do what you come to a nicely framed 4x3 filling shot? Make a mess, that's what. :(

With SD, people are free to zoom and crop, stretch, or keep the OAR, as they choose on their TVs. So give them 4x3 and let them mess it up themselves. With HD via HDMI, these format switching facilities are often absent, so they'll have to watch what you provide.


The obvious answer (!) is to ask BigDog660 what they want: black bars, or cropped footage.

Cheers,
David.

[tedkunich]

And now for the questions:

- Keeping in mind my originals are 1024 x 768 already, what HD format should I go for?
- Would it not be better to let the player or TV do the upscaling?



I have had very good results with pillar boxing the results to play on my 50" screen - so far I have played just the avi file through my media player, not burned to DVD. I prefer the original aspect ratio vs a cropped image.

T

[SeeMoreDigital]

Standard 8mm film offers a visible size of 4.8 x 3.5mm, which equates to an movie aspect ratio of 1.371:1. This is slightly wider than 1.333:1 (ie: 4:3).

So if you were capturing at 768 pixels high, your width should have been 1056 (1053 is the exact) pixels, instead of 1024 ;)

[Alex-Kid]

My choice is: 720p24 pillar boxed

You need to add 1 frame for every 3 to get 24 fps. It can be done by blending (ConvertFPS, maybe not desired) or by interpolation (MVFlowFPS). The latter may introduce some artifacts, but I think you must give it a try (specially with MVTools2 → MFlowFPS at slowest/highest quality setting). You will lose "originality" (just 1 of 4 frames will be same as source) but will eventually gain smoothiness and preserve film look.

18 fps footage is harder to work, huh!

[2Bdecided]

videoFred,

Can you post a clip? Rather than talk about it, I'd like to try the various frame rate options.

Cheers,
David.

[videoFred]

I will post a clip as soon as possible, David. I do this in my free time and I have lots of other work to do now.

Thank you for the interest in my work! ;)

Fred.

[2Bdecided]

Hey, ditto, no rush! :)

[*.mp4 guy]

It should be possible to encode at 16fps (or close to it) and get the "correct" frame rate via soft pulldown flags, like how 3:2 pulldown is handled.

[bigdog660]

Hey Guys,

While cropping would look quite nice on wide screen monitors, I am afraid too much of the picture would be lost. Given it's amateur film and the camera is all over the place (off center from the subject), I think pillarbox is the best.

You lose content, and in this case you lose resolution (and you have none to spare!). You'd also have to do it manually - just because the interesting part is that the bottom of this frame doesn't mean it stays there - in the next shot the same cropping will probably remove everyone's head.

Exactly! This is also why I prefer 1080p over 720p.;)

The was some mention of 50i, but I'm in the States, so I guess we need to make that 60i; however, I would like to avoid interlacing like the plague if we can. These films will be authored to BD (and resized and authored to DVD), so I believe I need 23.976, not 24.00, 25.00 or 29.976 fps. Please correct me if I'm wrong, but isn't 24.00 fps used in AVCHD only? 25.00 is pal, and 29.976 is telecine (I'm also wanting to avoid this).

@SeeMoreDigital, Isn't true though that BD has to be 1920x1080 or 1280x720, so we can not use an "oddball" frame size? Using your calculation, we'd need to upsize to 1480x1080 to keep 1.371:1. I guess it could be done with pillarbox?

The only problem with your calculations is videoFred's camera maxes out at 1024x768, so there is no way to keep the 1.371:1 aspect ratio unless he can capture at 1024x746 and then upsize at a ratio of 1.45 to obtain 1480x1080.

@videoFred, from what I've searched on the web, the original film rate is 18fps.

[smok3]

forget 50 anything, i was reading the original post wrongly.

Upsizing 4:4:4 material shouldn't be such a problem, what 'we' have is a good source, so maybe really full-res makes sense here.

one could pillarbox blowup like this;
http://resizecalc.somestuff.org/index.php?ssmw=1024&sar=1&sar2=&ssmh=768&CT=&CL=&CR=&CB=&mplayCrop=&trw=1440&dar=1&dar2=&modw=&modh=&padw=1920&padh=1080&css=&doit=true

[videoFred]

The only problem with your calculations is videoFred's camera maxes out at 1024x768, so there is no way to keep the 1.371:1 aspect ratio unless he can capture at 1024x746

I can do that, I can set the camera to everything below 1024x768, with the ROI (region of interest).


@videoFred, from what I've searched on the web, the original film rate is 18fps.

Let's say it will be close (clockwork camera's) :D

Fred.

[videoFred]

Upsizing 4:4:4 material shouldn't be such a problem, what 'we' have is a good source, so maybe really full-res makes sense here.

The original output from the camera is Y800 not debayered, debayering is done on the fly by the capture software, then saved to uncompressed RGB24 AVI.

Fred.

[EuropeanMan]

My god...what genius this guy is (videoFred)

Just in awe in all of this....

PS - what is pillarboxing? Is it just simple lanczos4resizing?

[SeeMoreDigital]

Using your calculation, we'd need to upsize to 1480x1080 to keep 1.371:1. I guess it could be done with pillarbox?Indeed generating a "pillarbox" style encode at 1.371:1 is what would need to be done... Perhaps something like this: -

http://i40.tinypic.com/2hgwtao.jpg


Cheers

[2Bdecided]

This is also why I prefer 1080p over 720p.;)But they'll be no more "real" resolution, and it'll prevent you from using 60p.


On another point raised: I've no idea if BluRay supports arbitrary soft pulldown flags. I see no reason that it wouldn't, but then it does 24p properly, so doesn't need them in the same way as DVD did.

Cheers,
David.

[bigdog660]

But they'll be no more "real" resolution, and it'll prevent you from using 60p.


On another point raised: I've no idea if BluRay supports arbitrary soft pulldown flags. I see no reason that it wouldn't, but then it does 24p properly, so doesn't need them in the same way as DVD did.

Cheers,
David.

Wouldn't there be a small loss since I'd be resizing from 1024x746 to ~986x720?

Okay, this is where I get confused... when using eac3to, I've seen 24p/1.001 and 60i/1.001. I have never seen a BD with 60p. Is there such a thing?

If there is, I'd just use an avisynth script to get 23.976 and resize to 480p (for a DVD copy)?

@anyone videoFred and I also need advise for a resize script that upscales to 1080p and downscales to 480p for DVD. Any suggestions are welcome. I've heard some good things about lanczos4resizing and Spline36Resize, but want to avoid ringing.

Thanks,
bigdog660

[videoFred]

Hi BigDog,

My filmscript contains LimitedSharpenFaster(). It is using internal upscaling/resizing already so we can do the resizing with it.

It would be a good idea to make the different versions from the same source: once upscaled, once downscaled.

Now, if I look at my standard PAL 720x576 files on my Dell 1440x900 monitor (pillarboxed of cource) I must say it looks very good. (with MediaPlayerClassic).

So I still think there is no need for 1080p. But we can test all options of cource.

Fred.

[smok3]

you would need to test encoded 1080p vs encoded 720p (with 1080p you would gain some chroma resolution and probably loose some with the encoder/limited bitrate, dunno...)

[2Bdecided]

Okay, this is where I get confused... when using eac3to, I've seen 24p/1.001 and 60i/1.001. I have never seen a BD with 60p. Is there such a thing?Yes, but I meant 60/1.001 - "60p" is lazy shorthand for those who know that no one really uses 60.000p.

Whether you lose resolution with 720p vs 1080p depends on how much resolution you have to start with. It's got little to do with the number of pixels you start with - it depends how sharp they are.

If the source is pixel-sharp 4:4:4 then there's an obvious advantage to using 1080p - however, it looks to me like it's nowhere near pixel sharp - so the question becomes whether the sharpening gives something that looks better at 1080p, or is already fully represented by 720p.

Cheers,
David.

[onesloth]

I've seen 24p/1.001 and 60i/1.001. I have never seen a BD with 60p.

http://en.wikipedia.org/wiki/Bluray#Technical_specifications

[Sagekilla]

Yes, but I meant 60/1.001 - "60p" is lazy shorthand for those who know that no one really uses 60.000p.

Whether you lose resolution with 720p vs 1080p depends on how much resolution you have to start with. It's got little to do with the number of pixels you start with - it depends how sharp they are.

If the source is pixel-sharp 4:4:4 then there's an obvious advantage to using 1080p - however, it looks to me like it's nowhere near pixel sharp - so the question becomes whether the sharpening gives something that looks better at 1080p, or is already fully represented by 720p.

Cheers,
David.

The other thing is whether his imaging device produced full RGB pixels. Most sensors out there capture in a bayer pattern and the real resolution is somewhere in: (Height * width) / 4 < actual Res < Width * height.

Assuming he got a good imaging device, then I wouldn't doubt it's possible to get 1024x768 pixels of resolution out of most film formats.

[JohannesL]

A decent encoder isn't going to waste bits encoding the duplicate frames
Is x264 not a decent encoder? 30p > 60p makes quite a difference in bitrate here, with max refs.

[bigdog660]

http://en.wikipedia.org/wiki/Bluray#Technical_specifications

I see. I guess I've never seen it since all I have ever ran into on BD is 1080p24 and 480i60. So this opens another option to cap at 720p60. Thanks for the link.

If we were to use 720p60, is a avisynth script possible to covert RGB24 to 480p24 (for DVD)?

Is it worth capping at 60p over 24p when the source was 18fps to begin with? I guess the motion supposedly would be more fluid?

Is the source detailed enough to recommend 1080p24 over 720p60? I suspect it might be in some scenes and in others perhaps not. I'm sure it depends on the film and the camera's settings (namely focus).

I'm just trying to find the best solution since the capped film will be archived to HDD, authored to BD and authored to DVD (for family members w/o a BD player).

Perhaps when videoFred has some time (there's no rush), he can provide a pillar boxed sample at 1080p24 and 720p60 so we can compare. His current script seems to work wonders as seen on page 1. Another question is even if the source is worthy only of 720p, does videoFred's shapening script work enough magic to make the processed video worth going to 1080p?

@2Bdecided: Yes, I knew 60p=60p/1.001=59.976, I just didn't know BD supported 60p at all.

@JohannesL: I plan to encode the cap using x264 no matter what frame rate and resolution we decide (except for DVD of course).

@videoFred: I've ordered the external 1TB HDD.

bigdog

[Comatose]

Hey videoFred, do you mind sharing the script you used for the screens in the OP? I'm curious to see how you got to that end result.

[2Bdecided]

Fred's script has an entire thread it itself - keep up Comatose! ;)

@bigdog660,

18fps defines the smoothness of motion. Integer multiples of that rate will allow the same limited smoothness to be preserved. Non-integer multiples will introduce additional stutter because some original frames will be repeated more (i.e. displayed for longer) than others.

Assuming a 16fps source, if you choose 24p, some frames will be displayed for twice as long as others, a 100% error in duration, occurring 9 times per second. If you choose 60p, some frames will be displayed for 1/4 shorter than others, a 25% error in duration, occurring 4 times per second.

You can assume a different source (15fps, 17fps, 18fps etc etc) and get slightly different numbers. 60fps is an integer multiple of 15fps, for example.

Which is more annoying - a bigger stutter, and infrequent stutter, or a slight speed change? You'll have to watch the result to find out - though I but the subjective impression might even be different between NTSC users (who are used to 3-2 stutter) and PAL users (who are used to speed up, but no stutter)

Cheers,
David.

[Comatose]

Oh :P I guess I haven't been trolling enough lately :p

[bigdog660]

Fred's script has an entire thread it itself - keep up Comatose! ;)

@bigdog660,

18fps defines the smoothness of motion. Integer multiples of that rate will allow the same limited smoothness to be preserved. Non-integer multiples will introduce additional stutter because some original frames will be repeated more (i.e. displayed for longer) than others.

Assuming a 16fps source, if you choose 24p, some frames will be displayed for twice as long as others, a 100% error in duration, occurring 9 times per second. If you choose 60p, some frames will be displayed for 1/4 shorter than others, a 25% error in duration, occurring 4 times per second.

You can assume a different source (15fps, 17fps, 18fps etc etc) and get slightly different numbers. 60fps is an integer multiple of 15fps, for example.

Which is more annoying - a bigger stutter, and infrequent stutter, or a slight speed change? You'll have to watch the result to find out - though I but the subjective impression might even be different between NTSC users (who are used to 3-2 stutter) and PAL users (who are used to speed up, but no stutter)

Cheers,
David.

I think Fred was considering interpolation from 18fps to 23.976fps. If I'm not mistaken, this should cure the stuttering problem, but may leave some artifacts. If he uses interpolation, is there any benefit going 18fps --> 59.976fps?

I'm obviously out of my depth here, so please bear with all my questions.

Thanks,
bigdog

[videoFred]

The other thing is whether his imaging device produced full RGB pixels.

No, it does not. It's producing a Bayer mozaic(and thus B/W)picture in 1024 x 768. This picture arrives in the computer and the software is doing the debayering on the fly. The benefit of this is 3x faster fire wire transfer speed and better software debayering (edge sensing).


Assuming he got a good imaging device, then I wouldn't doubt it's possible to get 1024x768 pixels of resolution out of most film formats.

My imaging device is OK. I have tested it against almost everything else available. :) It is capable to show the limits of the old 8mm format. A very expensive 16 bits high-end device will not show more picture information then what I can show with my unit. (on this specific footage).

Of cource I agree upscaling and post processing will be easyer with a better source to start with. But with some smart scripting the final -visual- result will be nearly the same.

PS: I have added a new clip on Vimeo to show the effect of MVDegrainMulti().... Averaging 6 frames.... It's pretty spectacular. :)

Fred.

[videoFred]

I have made a first testclip: 720p. (pillarboxed) No interpolation, just an "assumefps(18)" followed by "changeFPS(60)", and thus lots of repeated frames.

Guess what: it plays remarkable smooth at 60fps on my Dell with MplayerClassic and ffdshow. Ffdshow can easily produce 60fps with nearly no CPU load. The format is simple: Microsofts Mpeg4V3. (fourcc MP43).

I'm happy :)

Fred.

[Sagekilla]

No, it does not. It's producing a Bayer mozaic(and thus B/W)picture in 1024 x 768. This picture arrives in the computer and the software is doing the debayering on the fly. The benefit of this is 3x faster fire wire transfer speed and better software debayering (edge sensing).


You're missing my point. In a bayer sensor, each photosite is only one color. You have ~200k red sensors, ~200k blue sensors, and ~400k green sensors. Interpolating those values to get 1024x768 resolution isn't perfect so you're not getting the full 1024x768: You can come close but you can't get it perfectly. a 3 CCD would be able to reproduce the image very sharply, whereas the bayer would look softer.

Yes, there are very good debayering algorithms that can make -very- sharp images, but they're all creating data they couldn't capture because it's based off a bayer sensor.

[videoFred]

You're missing my point.

I was not missing your point. I was answering to this:
The other thing is whether his imaging device produced full RGB pixels.


so you're not getting the full 1024x768: You can come close but you can't get it perfectly.

Yes, this is true. My telecine friends are working on a triple RGB capturing system with top quality high rez. B/W cameras. But for 8mm this is overkill.


Fred.

[Mug Funky]

@ videofred: try the approach the Arriscan (and i think the Vialta) take. have 1 black and white camera, and use LEDs in red green and blue. pin-register each frame and shoot it with red green and blue in sequence.

another advantage is that you can "double-flash" at different intensities (you can control the brightness of the LEDs and calibrate them to your film base), and make a high dynamic range image out of 2 low ones. this can give you much more latitude, which you'll need for all that reversal or print stuff you'd be getting.

one problem is when a piece of dust gets dislodged and flies around on the film's static charge - you'll see red green and blue confetti on some frames if this happens. it's only a problem at high scanning speeds though.

[videoFred]

have 1 black and white camera, and use LEDs in red green and blue. pin-register each frame and shoot it with red green and blue in sequence.

Yes, this is what my telecine friends are preparing right now. The RGB Led flashing unit is in full development. The resulted AVI file will contain a sequence of 3 RGB frames for every film frame. Selectevery() and MergeRGB() will be used for generating the colored image.


Fred.

[*.mp4 guy]

How are you getting around the "green gap" in the LED light spectrum, AFAIK there are no good green LEDs, only blue and red. Which is why LED based lights look purple-ish unless they are treated with light-frequecy-converting fluorescents. Are there green LEDs that are just unviably expensive for use as part of light fixtures?

[videoFred]

How are you getting around the "green gap" in the LED light spectrum,

I am using cool white Luxeon Star leds for now, but I'm looking forwards for an RGB Led setup.

This is what my telenice friend says about the Led spectrum "green gap":

the cool white LED you are using has a spike at blue and yellow, and a low at green and dark red. this is why you get higher saturation then with filtered halogen, but also more problems with skin tones.

a true RGB LED solution will have some gaps too, but with careful profiling it will have a very good range and less crosstalk then a filter solution.

Fred.