Optimizing VHS capture with current equipment

[JoePoe]

Hello!

First, these are my goals:
-With my current equipment, get the best possible digital backup of my VHS collection, by
-capturing the 288 line fields of my 576i content
-performing any post processing offline on these raw captures

Now, my equipment is this (happen to have, not bought for this purpose):
-Sony VCR SLV-SF90
-RetroTink 5x
-Elgato HD60s
-VirtualDub 1.10.4 64bit

What I did so far:
-Set RetroTink to output 1200p, with Bob deinterlace and V-sync set to TBC
-Capture at 50fps, 1920x1200, YUY2
-Load the video into cv2 to confirm that each frame contains 288 unique lines, repeated 4x after cropping black filler lines
(bob deinterlace fills the frame by doubling lines, and scaling to 1200p means 2x integer scaling with some zero padding in this case)
-extracting each field with frame[24:1176:4,:,:] and writing to a new video with 50fps and 288x1920 resolution

This produces video that looks good with an image that shifts up/down when frame stepping. However, applying any sort of deinterlacing filter, significantly worsens the image which leads me to believe I haven't properly stored it as interlaced video.

So my first question is, what is the proper way of storing the extracted fields, to recreate a 576i video? How do I make sure I do not mess with the color encoding?

My second question regards the horizontal resolution. My current understanding is that there is no defined resolution horizontally, but rather a continuous analog signal. Then my thought is that this is best preserved by sampling at a high frequency (i.e. high resolution). With the current method, I sample each line 1440 times (the rest is zero padding to a 16:9 frame). Let me know if I got this backwards in some way!

I'm sorry if this got lengthy, but I appreciate any answers!

[lordsmurf]

You lack any form of TBC. That will not work. TBCs are not optional.

Retrotink is for video games, and will not be useful for videotapes. But have "video" in the name, but any random video is another random video.

Elgato earned the nickname Elcrapo.

At bare minimum, you must add an ES10/15 as minimalist TBC(ish), and get a capture card made for analog video capture that isn't junk.

[JoePoe]

You lack any form of TBC. That will not work. TBCs are not optional.

Retrotink is for video games, and will not be useful for videotapes. But have "video" in the name, but any random video is another random video.

Elgato earned the nickname Elcrapo.

At bare minimum, you must add an ES10/15 as minimalist TBC(ish), and get a capture card made for analog video capture that isn't junk.

Thank you for answering. I'm afraid you did not answer any of my questions though. I tried stating my goal to avoid this.

However, I did mention that v-sync was set to TBC, an option that was added to recent firmware in retrotink. Regarding Elgato, I can't really see that it does any heavy lifting since it basically just transfers a digital signal, but still, my prerequisite here is to work with what I have.

My main question is still, given my capture that I detailed, and given the extracted fields, how do I properly reencode them into a video file?

[Bwaak]

Why 1200p? Don't both the Retrotink and Elgato output 1080p? You don't need to deinterlace. If you've got 50p, you can keep it like this. Adjust aspect ratio and frame size as needed.

Why do you want to recreate 576i?

SD digital video has 720 samples per line in active area. SD analog video is usually sampled into 720 samples per line per Rec 601. 1440 is overkill.

With your workflow, I would keep the resulting clips in 1440x1080p50 - watchable on a computer, uploadable to YT.

[JoePoe]

Why 1200p? Don't both the Retrotink and Elgato output 1080p?

1200p was the only output resolution that preserved all visible content. It uses 2x integer scaling with zero padding (576*2+48black lines). 1080p from this device will either integer scale with some cropping or use non-integer scaling which adds undesirable interpolation. It can output 480i but not 576i so this was my best option.

You don't need to deinterlace. If you've got 50p, you can keep it like this. Adjust aspect ratio and frame size as needed.
Why do you want to recreate 576i?

I wanted to capture the original fields, since my understanding is that you can get much higher quality deinterlacing in post processing. Also, my philosophy is that keeping captures with minimal processing applied to them, is better for long term preservation (you can go back to raws and redo the post processing at any time in the future).

SD digital video has 720 samples per line in active area. SD analog video is usually sampled into 720 samples per line per Rec 601. 1440 is overkill.

Thank you, this is reassuring. I guess it is difficult to ensure that your sampling is perfectly aligned with the original sampling though? And you would have added noise as well, so I'm happy with a bit of oversampling then.

With your workflow, I would keep the resulting clips in 1440x1080p50 - watchable on a computer, uploadable to YT.

My idea is that "watchable/uploadable" comes in the post processing step, this more concerns the raw captures.

[Bwaak]

my philosophy is that keeping captures with minimal processing applied to them, is better for long term preservation

I don't think this is possible with your current hardware.

[lordsmurf]

I tried stating my goal to avoid this.

Your goal cannot be achieved with your current gear. Quality will suck, the experience will be miserable and full of issues. You're using the wrong tools, and lack of tools, the end, no further room for discussion.

However, I did mention that v-sync was set to TBC, an option that was added to recent firmware in retrotink.

That's not a TBC.

My main question is still, given my capture that I detailed, and given the extracted fields, how do I properly reencode them into a video file?

You're SOL.

[dellsam34]

If you have a PAL video you're wasting your time, Get a proper capture device for the right format that can capture lossless 720x576 interlaced, and proceed from there.

[JoePoe]

I've been quiet for a while now, since the feedback I got here made me rethink things. In the end I realize that I can get massively better quality backups by using a professional service than doing it myself.
I found a local service that instills a lot of confidence judging by their equipment and detailed documentation on saving difficult tapes. The only thing that is not well described is the digitization part. So I wanted to check with you what requests regarding backup format I should make before turning in the tapes?
My understanding is that preserving the original video format w/o upscaling/deinterlacing is preferrable while also using a lossless compression. Would you agree, or do you think that they do a better job in that department than I could ever hope for anyway?

[Sharc]

The most critical phase in the process is the capturing of the tapes using qualified equipment. No idea how competent your "professional service" really is and how they are doing it, how much time they spend on it etc. I would give them just 1 tape (with mixed scenes, dark, bright, motion, noise) to begin with and see what they deliver. Post the result here to receive comments from members.

I would recommend to keep the tapes plus a decent interlaced, lossless - or at least visually lossless - unprocessed captured version in your archive. If you should come to new conclusions or find a better service this will be your repository to start with for future work. Just in case.

[dellsam34]

Could you share the name of business? As far as I know most of them are geared towards capturing directly into a compressed format, so the option of having lossless files is non existent to begin with.

[JoePoe]

Could you share the name of business? As far as I know most of them are geared towards capturing directly into a compressed format, so the option of having lossless files is non existent to begin with.

It's in Swedish but here's their links:
https://www.raddabanden.nu/
https://www.facebook.com/raddabanden/

[dellsam34]

They appear to have some professional gear, at least better than using a VCR/DVD combo and elgato stick that most businesses use.

[Sharc]

They seem to charge about 25.- USD for the 1st cassette, less for higher quantities. So it's worth to try IMO. Technically they should be able to provide lossless or visually lossless captures as well if they are willing to. Just ask them for the conditions and delivery formats.
Lossless / near lossless captures could serve as your masters for later processing with an NLE (cutting, joining, transitions, adding titles and audio .... etc).

[JoePoe]

I was just on the phone with them. Apparently they do archival work for SVT (our equivalent of BBC).
They always deliver interlaced sd format.
They can provide completely uncompressed video in "8b YCbCr" (which I guess is another way of saying yuv422?). That is 70gb per hour.
They do not provide huffyuv encoding since that is not available in their software.
They recommend ProRes 422 or 4444 for visually lossless.

Now, let's say I go with uncompressed and reencode to lossless format, what compression is preferred and what gain in file size can I expect? How does that compare to ProRes?

[Sharc]

Sounds reasonable.
Not quite sure what they understand by "completely uncompressed video in 8b YCbCr". Maybe 8bit YUV 422, 8bit YUV444, or raw? What decoder would you need? What tools do you have?
Any 444(4) color subsampling format is way overkill for VHS or S-VHS captures, 422 is just what you would need.
ProRes422 (.mov) is fine as a visually lossless intermediate format supported by very most NLEs AFAIK.
It depends what you intend to do with the captures (post precessing, tools ...).

Now, let's say I go with uncompressed and reencode to lossless format, what compression is preferred and what gain in file size can I expect? How does that compare to ProRes?

- uncompressed 8bit YCbCr -> Huffyuv or similar: you can expect a reduction of 2....3, so typically around 20 ...30GB per hour (720x576,25fps).
- ProRes 422 is 18GB per hour (720x576,25fps).

Audio comes on top.

[JoePoe]

Sounds reasonable.
Not quite sure what they understand by "completely uncompressed video in 8b YCbCr". Maybe 8bit YUV 422, 8bit YUV444, or raw? What decoder would you need? What tools do you have?
Any 444(4) color subsampling format is way overkill for VHS or S-VHS captures, 422 is just what you would need.
ProRes422 (.mov) is fine as a visually lossless intermediate format supported by very most NLEs AFAIK.
It depends what you intend to do with the captures (post precessing, tools ...).

Now, let's say I go with uncompressed and reencode to lossless format, what compression is preferred and what gain in file size can I expect? How does that compare to ProRes?

- uncompressed 8bit YCbCr -> Huffyuv or similar: you can expect a reduction of 2....3, so typically around 20 ...30GB per hour (720x576,25fps).
- ProRes 422 is 18GB per hour (720x576,25fps).

Audio comes on top.

Thanks!
The gain in size does not seem big enough then that it would argue for going lossy. (I know this is mostly my OCD speaking though. )
I need to double check how the uncompressed video would be encoded, that was not entirely clear to me.

In any case, to be able to match what they do I would need to spend huge amounts of money on equipment alone and my impression is that I would still do a worse job given their experience/knowledge so going this route seems like a no-brainer.

Thanks to all for valuable input!

[Sharc]

I need to double check how the uncompressed video would be encoded, that was not entirely clear to me.

Yes, ask them about the '8b YCbCr' details. It is possibly YUV4MPEG2 with the extension .y4m. Or perhaps just raw .yuv? It can be 444, 422 or 420. Basically it can be opened with ffmpeg but there seem to be some uncertainty about the color matrix and some questions with ffmpeg regarding the interpretation of the chroma placement. Maybe this has been solved in the meantime, I don't know. For SD material the chroma placement it is not very critical though, AFAIK. Also, how will they deliver the audio?
https://wiki.multimedia.cx/index.php/YUV4MPEG2
Maybe ProRes422 is the better choice for your purpose.

Perhaps someone more knowledgable will chime in .....

[pandy]

Strange that they not offer 8.2 format as standard for broadcast equipment - in PC world 8.2 this will be like yuv420p10be, yuv420p10le, yuv422p10be, yuv422p10le, yuv444p10be, yuv444p10le.
Probably Prores is more than enough and way easier to manage than RAW YUV.

[JoePoe]

I got some more info on the lossless format that he provides. Apparently, he uses software that comes with the AJA card used, and that lists a few "built in" formats:
10 bit BE RGB
10 bit YCbCr
12 bit RGB
8 bit BGRA
8 bit YCbCr

I guess it is Mac based software since they are all in mov containers? Maybe this gives a clue to someone what the specifics of "8 bit YCbCr" is.
He only works with lossy codecs (like prores) himself so can't give more details.

Maybe it is easiest to go with Prores 422 HQ?

[themaster1]

Prores 422 hq is just what you need.