Re-encode old captures from DVD recorder to future proof?

[ido2]

Hi all,

Maybe a question which would interest others as well:

For a few dozen Video8 + VHS-C PAL tapes, captured w/TBC-1000 + JVC DR-M10, then ripped to PC and merged to a single file, kept as MPEG2, no editing or processing done.

I also have a USB live2 dongle, which I haven't used because I decided back then not to bother with calibrating histograms (and etc) for capturing and restoration.

1. If captured at XP quality, how bad would it be if I would deinterlace, re-encode for viewing on PC/streaming ?
- If not so bad, how should I proceed with the conversion (workflow-wise from mpeg; there are enough threads on how to use virtualdub with various .avi containers)

2. I have some new tapes to capture - should I keep using my current workflow or bother with the USB Live?

Thanks in advance,

[latreche34]

MPEG-2 is an ancient non efficient lossy format, anything you do to it will degrade it further, The best way is to recapture the tapes to lossless AVI and proceed from there, If the tapes are gone and cannot be recaptured the best thing you do is just leave the MPEG-2 files alone.

[ido2]

latreche,

I know MPEG-2 is a lossy format, however, the source material is ONLY around ~240 lines of resolution captured at XP, which is 720 x 480 @ 9 Mbps.

Although this is not what @Lordsmurf considers superbit (at 15 Mpbs), which retains all the original data, but in my above question I was wondering if anyone tried this conversion, and if 9 Mpbs has enough data is retained in the video file to consider good enough quality for conversion.

If I'll go the recapture route, I need to do some reading and brush up on recapturing with the USB-Live2 (should I capture at 720x576 or 360x576, after all source is ~240 lines..)

[latreche34]

The hardware captures natively @ 720x576 (PAL) regardless what resolution you choose and regardless of what you think the quality of the source might be, Any resolution you choose other than 720x576 the driver will resize to it on the fly during capturing, This is well known as a bad practice in the capturing community as it is part of the post tasks such as de-interlacing, upscaling, color correction ... etc.

As to your MPEG-2 files, honestly they are of a very low quality to do anything to them in my opinion, I let other members help you on that path.

[themaster1]

Minimum you can do is 704*576 or 480 if ntsc imo which is dvd compliant
9000 kbps is good enough if :
1-you have a good "professional"encoder (not all are the same ). Mainconcept reference is the best for me (kinda old)
2- it's a software 2 pass encoding. On the fly mpeg2 encoding vary from bad to medium quality at best (macroblocking etc..) with this bitrate.

[Hushpower]

Quote:

Originally Posted by Latreche

The hardware captures natively @ 720x576 (PAL) regardless what resolution you choose and regardless of what you think the quality of the source might be, Any resolution you choose other than 720x576 the driver will resize to it on the fly during capturing, This is well known as a bad practice in the capturing community as it is part of the post tasks such as de-interlacing, upscaling, color correction ... etc.

Can you clarify what you mean? What is bad practice: capturing at 720x576 or forcing the capture driver resize on the fly?

[latreche34]

Means resizing on the fly to a different resolution other than the native hardware sampling.

[traal]

Quote:

Originally Posted by ido2

(should I capture at 720x576 or 360x576, after all source is ~240 lines..)

240 lines per picture height is equal to 320 pixels across on a 4:3 screen (240x4/3). Nyquist says you should at least double this for capture, so you need at least 640 pixels across to get all the information. For PAL, this means 640x576 or 720x576 or 768x576, whatever your capture card supports.

[lordsmurf]

Quote:

Originally Posted by ido2

For a few dozen Video8 + VHS-C PAL tapes, captured w/TBC-1000 + JVC DR-M10, then ripped to PC and merged to a single file, kept as MPEG2, no editing or processing done.

To start, good hardware, so conversions should be fine if no more than 4 hours per disc (Full D1 1-2, Half D1 3-4, which decent bitrates).

Quote:

1. If captured at XP quality, how bad would it be if I would deinterlace, re-encode for viewing on PC/streaming ?
- If not so bad, how should I proceed with the conversion (workflow-wise from mpeg; there are enough threads on how to use virtualdub with various .avi containers)

If XP mode, not really that bad. It can be viable. Not best, due to MPEG limits already discussed here, but viable. I've done it with my hobby recordings. I cannot capture everything lossless (mostly 90s/00s TV recordings), nor does it necessarily deserve to be.

Open in Hybrid, QTGMC deinterlace, crop 16 then resize AR to 4x3, encode H.264 with mid/high teens CRF. All DVD recorders messed with IRE some, including the best JVCs, so it is often ideal to tweak the black levels in the filter/Avisynth/Vapoursynth settings in Hybrid. (And as I always mention, that freeware is worthy of a small donation to the dev, selur. Always remember to give back some dollars to sites and freeware that help you on your journey!)

Quote:

2. I have some new tapes to capture - should I keep using my current workflow or bother with the USB Live?
Thanks in advance,

I think your current path is fine. I see nothing that needs change. Yes, it could be better, but I will never fault JVC LSI units, used with DataVideo TBC, with JVC VCRs with line TBC. If no quality VCR was used, then maybe we need to see a sample, as neither that DataVideo nor the JVC recorder do anything for line timing wiggle. And that unstable line timing can really screw with the JVC LSI encode quality, as well as latter H.264 encode/filter/deinterlace quality.

Quote:

Originally Posted by latreche34

MPEG-2 is an ancient non efficient lossy format, anything you do to it will degrade it further, The best way is to recapture the tapes to lossless AVI and proceed from there, If the tapes are gone and cannot be recaptured the best thing you do is just leave the MPEG-2 files alone.

While I agree in general, I don't in all cases, where certain quality gear is used. Pending VCR info here, I think he'll be fine.

Quote:

Originally Posted by ido2

latreche,
I know MPEG-2 is a lossy format, however, the source material is ONLY around ~240 lines of resolution captured at XP, which is 720 x 480 @ 9 Mbps.
Although this is not what @Lordsmurf considers superbit (at 15 Mpbs), which retains all the original data, but in my above question I was wondering if anyone tried this conversion, and if 9 Mpbs has enough data is retained in the video file to consider good enough quality for conversion.
If I'll go the recapture route, I need to do some reading and brush up on recapturing with the USB-Live2 (should I capture at 720x576 or 360x576, after all source is ~240 lines..)

360x576 is not valid, 352x576 is the Half D1 PAL resolution.

As per above, I run that method with NTSC hobby tapes. It looks fine. Especially if you run a 4:2:0 H.264 encode (from lossless source) instead of 4:2:2, as DVD is already 4:2:0. (Read that carefully, don't misread, I'm NOT saying you should convert DVD sources to 4:2:2 H.264 in this scenario.)

Quote:

Originally Posted by themaster1

Mainconcept reference is the best for me (kinda old)

Off-topic: It's still current software, with a current version, now known as MainConcept TotalCode Studio. I got OS+hardware locked a while back (and don't want to switch seats), can no longer update. But those are available. $500 is a bit high, but it's also not gone up with inflation, so there's that? For many encode tasks, there is no alternative, not the reverse engineered x264, not the reverse engineered ffmpeg. Pro needs, pro software.

I rarely recommended this software, for that reason. Non-pros can subsist on x264/ffmpeg with GUIs. Now if we could only convince selur to add MPEG encoding to Hybrid! The main limited to ffmpeg GUIs is lack of 4:2:2 encoding at 15-50 superbit.

Quote:

Originally Posted by latreche34

Means resizing on the fly to a different resolution other than the native hardware sampling.

Hardware encodes do this much better than capture cards, excluding ATI AIW in MPEG (not AVI/lossless/raw) due to the Ligos hybrid hardware assist. These scale well, not stupid rough abusive downconverts, like most all capture cards.

Quote:

Originally Posted by traal

240 lines per picture height is equal to 320 pixels across on a 4:3 screen (240x4/3). Nyquist says you should at least double this for capture, so you need at least 640 pixels across to get all the information. For PAL, this means 640x576 or 720x576 or 768x576, whatever your capture card supports.

For usage, I've never been fond of Nyquist. It's raw, unrealistic, and doesn't consider practical card application. It's good info for engineers/devs making cards, fairly useless to those using cards available. At best, it helps select a non-POS capture card.

[lollo2]

Quote:

For usage, I've never been fond of Nyquist. It's raw, unrealistic, and doesn't consider practical card application.

Non sense.

PAL signal is 576 lines, each has 5MHz bandwidth, then 384 max resolution for every line. Nyquist–Shannon sampling theorem will give then 384x2=768, defining the PAL digital frame. Because the resolution/bandwith is never that high, the choice was to reduce to 720x576 / 702x576. That's the basic of digital video and the its origin.

[lordsmurf]

Quote:

Originally Posted by lollo2

Non sense.
PAL signal is 576 lines, each has 5MHz bandwidth, then 384 max resolution for every line. Nyquist–Shannon sampling theorem will give then 384x2=768, defining the PAL digital frame. Because the resolution/bandwith is never that high, the choice was to reduce to 720x576 / 702x576. That's the basic of digital video and the its origin.

I had these sorts of conversations 20+ years ago. You have Nyquist sticklers, and you have Nyquist pragmatists. Yes, yes, I know the math, the theory, it's all sound. But it just isn't overly useful to practical application where variables exist in the hardware.

I still remember arguments from folks who insisted VHS source resolution equivalency was more than 352x480, due to Nyquist, but it was (and still is) nonsense. I did a quick Google to find a conversation I recall, but instead found another conversation (on a site I'd never heard of), between two folks, on this topic, from 1999. The pro-Nyquist person was rigid, and the non-Nyquist person correctly pointed out that aspect such as sampling determine the resolutions. Add in concepts like oversharpening, variant VCRs, etc, and it's clear as mud.

Math is art.
Math lies.
Folks that don't realize this don't use it heavily. Numbers are always fungible. When you understand that, and understand the other video concepts, you see why Nyquist is only a theorem (not a constant).

[lollo2]

Quote:

VHS source resolution equivalency was more than 352x480

Wrong, because the bandwidth of the recorded lumincance in VHS. Nyquist is always right, even in this case.

Quote:

Math is art.
Math lies.

And this closes all discussion, at least on my side.

[lordsmurf]

Quote:

Originally Posted by lollo2

And this closes all discussion, at least on my side.

For the best.

As I said, age-old discussion, literally going back 20-30 years.
Kirk vs. Picard, Mac vs. PC, VHS vs. Betamax, etc.
Each side thinks they're 100% correct, but it's so much fuzzier than that. And that's always been my stance. The world isn't B&W, but lots of shades of gray.

After years, certain topics get grating. But then something funny happens. It becomes amusing to see such old disagreements still lingering.

Just remember: you and I agree more than disagree. And I rarely inject when you post about Avisynth, those topics I always find interesting.

Speaking of Avisynth, this OP in this topic may find your Avisynth-fu quite useful, he needs to post a new thread with samples.

[lollo2]

Quote:

Just remember: you and I agree more than disagree. And I rarely inject when you post about Avisynth, those topics I always find interesting.

Sure. It is always constructive, and a pleasure, to have discussion with you and the other experienced users

[latreche34]

Nyquist theorem describes the minimum sampling you can do to digitize a signal without loosing any original details and that applies to video, audio or any other type of analog signal that needs to be stored digitally, there is no argument below that limit. The filtering stages that come after the sampling is where the grey areas, such dithering and other noise shaping filters. The only arguments I heard is about oversampling or sampling higher than the Nyquist limit which can create aliasing and other unwanted artifacts according to them, especially for audio, a lot of digital audio enthusiasts believe that anything over red book is not good and they believe all this Hi-res and 24bit is all BS.

[lordsmurf]

Quote:

Originally Posted by latreche34

Nyquist theorem describes the minimum sampling you can do to digitize a signal without loosing any original details and that applies to video, audio or any other type of analog signal that needs to be stored digitally, there is no argument below that limit. The filtering stages that come after the sampling is where the grey areas, such dithering and other noise shaping filters. The only arguments I heard is about oversampling or sampling higher than the Nyquist limit which can create aliasing and other unwanted artifacts according to them, especially for audio, a lot of digital audio enthusiasts believe that anything over red book is not good and they believe all this Hi-res and 24bit is all BS.

I think I'm agreeing with you here...

The problem is in the rigid numbers. The values can be too high, or too low, compared to actual outputs from VCR into capture cards. The math is sound, but that's about it. However, other maths and measurements may also be valid for practical observations.

Nyquist is mostly useful for the engineers designing the components, and mostly worthless to end users. And yet, it's the end uses that bicker, usually for no reason. Why? Because we're not dealing with any sort of raw sources. We're dealing with VCR outputs, not signals. We've dealing with manufactured chipsets, not homebrew chips.

The vhs-decode project aims to extract data at the source, so I can understand needing to know Nyquist values. (Noting that project has limits and issues, it's just not working as well as anybody wants, aside from the apologists that see no faults. It may eventually become viable, or not. For now, it's very much a homebrew sort of setup.)

But for anybody using a VCR (as intended), you're not getting a "raw" signal, period. So pissing contests about Nyquist is quaint and amusing. Children on a playground. Usually measurebating resolutions. In past years, to try and poo-poo Half D1. In more recent times, to squabble about HD or audio. Sometimes it's simply being conflated with crappy quality resampling/scaling in cards/devices.

I just hate it when somebody insists 1+1=2, without exception, not understanding it's only true in number base 10. In base 2, aka binary, that would be nonsensical. (That's not the best example, but I forget how to add/subtract in base 8, it's been too long.) Where base changes get complex is when you understand that roundings are different, and can change formulas quite a bit. Sort of like sampling roundings.

Video is math, but math can be wibbly-wobbly.

[latreche34]

Whether the end user understands the theorem or not there is nothing can be done about it, It is a hardware design embedded in the ADC chips algorithm few decades ago, We cannot change that. So not only there is no gain in resizing bellow 720/704 during or after capturing, it actually may degrade the overall quality of the frame.

The vhsdecode I believe samples at a little over 900 pixels per scan line, not sure though and I'm pretty sure they have their own reasons, one of them being they capture the entire frame including the blanking area.

[lordsmurf]

Quote:

Originally Posted by latreche34

So not only there is no gain in resizing bellow 720/704 during or after capturing, it actually may degrade the overall quality of the frame.

It really depends on the quality of the sampling/scaling.

But even more salient is understanding why lower res existed: bandwidth savings, file size savings. These come from day long before multi-terabyte storage drives, and HD display/scaling/algorithms. So today, resolutions below 720 are moot, aside from using legacy DVD recorders in longer play/record modes.

So while academic disagreements of Nyquist/etc still happen, it's doubly pointless. 720 captures all analog formats with excess headroom, and all you need is a good card to properly capture values (exposure, luma, chomra, IRE, etc).

This feels too off-topic now.

The OP wanted to use his recorder to capture XP mode (720x480/576), then process+encode to streaming formats. What he's doing is fine, I approve, I do it myself for TV hobby work. I only re-capture value shows to lossless.