U-matic capturing compared, best Avisynth scripts?
 

Made a similar post over at videohelp since I thought they have a more active Umatic community, but maybe not, so though I'd post here as well since I'm now getting to some actual capture experiments.

Basically, I tried capturing the same U-Matic tape section in several different ways, though always saving as ProRes422 encoded by an AJA KiPro which should be roughly visually lossless and is more Mac friendly than lossless AVI for testing purposes. CVR samples are done over SDI and the DPS sample is via DUB and then sent over component to the KiPro - color saturation on that one is obviously more just due to how the DPS unit is set, so would try to ignore that as you compare. The KiPro will report in realtime if it is dropping frames and it did not report any dropped frames with any of these tests.

Here is a link to a folder that has the sample files:
https://www.dropbox.com/scl/fo/on1ml...ikfjl001q&dl=0

I apologize in advance for the annoying audio right at the beginning, it is on the tape like that. You'll want to actually download the files to see their full quality as the dropbox "player" plays a compressed version in most cases.

I also threw in an Elgato capture just to show how some YouTubers are capturing these tapes "professionally" which I guess that is technically true since that is their profession haha. I think they mean well, but I don't think they've really looked at how much better the capture quality could be with a few upgrades. The AJA is even Mac friendly and I'm sure just running that through handbrake would produce a pretty appealing result compared to the elgato for minimal extra steps and time.

Questions:

1. Is one chain/file obviously superior? The regular S-Video to me appears to have less ringing (which Umatic is known for), but at the cost of sharpness (or at least perceived sharpness/more grain). Not sure if Avisynth would do a better job of reducing the ringing post capture while preserving sharpness? I think the DPS and the pre-noise reduction S-Video look the best.

2. What specific Avisynth scripts would be recommended on these samples for QTGMC as well as to help get rid of some of the occasional line dropouts and ringing while trying to preserve as much sharpness as possible? If I were to upscale to say 1440x1080 for YouTube, is there a recommended scaler function that can also be added to the script?

FYI: VH is really more of a newbie community, with most topics being about (1) "rip"/copy discs, (2) download streaming. That is literally about 2/3rds (~66%) of the topics there. Then how to process in freeware or Chinaware. (Stuff like Hybrid is "too hard", and so they gravitate to inferior Handbrake/etc.) Most analog/capturing topics get input from members here, and then the Avisynth/restoration is mostly people that got sick of Doom9 (though 2020s Doom9 is vastly friendlier than 00s-10s). The main reason that VH even discussed capturing, especially restoration/filter, is because of my involvement, and my curation, back when I was a moderator there in the 00s.

Anyway...

No, you won't get much U-matic help there, unfortunately. But we do have quite a few members here with U-matic experience. My U-matic experience is post-capture, dealing with the issues that hardware cannot correct.

I'm not able to download your clips now, but I can comment based off your comments.

Ringing/halo is really distracting at a large size. Those lines can literally measure an inch thick on a large TV screen, so it's vastly better to have slight softness. We expect softness from old format, but not crayon outlines. (Halo/ringing often reminds me of 1st grade, where myself and a few other kids had this competitive style of hard inner coloring lines, with pastel shading. I still have a few samples in a binder, kept my childhood artwork.)

Dropout removal can have side effects, not too different from the butcher-job Topaz "upscaling". It has aberrations. So any dropout removal needs to be done at a micro scale. In other words, only segments with dropouts present, not the full video -- unless the full video has lots of dropouts. There's actually software to paint out issues, but it gets even more tedious and costly.

You really need to scale separate from restoration. If for no other reason than it just extends processing time. Very often, processing time is not additive. So 1+1 (two tasks together) = 3, and not 2. It's not additive, but rather bottlenecks and extends.

Since the captures have already been done, I think the same steps would apply post-capture as with VHS and I think and whether one capture chain produces a better result for next steps should hopefully be apparent to those that use avisynth regularly. Really it's just capturing a different NTSC source, but the same rules or signs of a better capture I think should still apply here.

VHS often can have ringing as well, but usually not quite to this degree.

And fair fair enough, I could see upscaling making more sense to do as a separate step.

With the Elgato you captured at 640 by 480 with a bitrate of 1,500kb/sec. That is a good comparison because with the Elgato people usually think of a “professional company” using it on a Mac. The Mac version is limited to 1,500kb/sec. In the Mac version of you capture 720 by 480 it disables deinterlacing.

With the windows version you can capture 720 by 480 at 11,300KB/sec. It still looks like crap but maybe not as bad.
https://youtu.be/G0zs_E1VETk?si=2dzkMCDMsE2BTj5c

I did get a couple of replies over on videohelp - Seems that clipping of luma is occurring in various places, could actually be at the CVR TBC's input. DPS475/575's are kind of known for that where if the input is over 115IRE, it clips at the input and adjusting with the built in proc amp just lowers the "already clipped" luma levels.

So needless to say, I do need to chase down if I have some settings wrong and I will get a waveform monitor in the capture chain since there's apparently nothing in the way of automatic gain control by default on these devices, or I just have something set wrong. There is a "legalizer" function within the CVR, but I believe that only scales the top end of luma rather than the whole signal (and from the manual, sounds like it clips anything below black to black without trying to scale that, but I'll do some more testing to see what it does with a known-hot ramp test pattern.

Testing how different AGC and legalizers behave is on my lists of tests to do, so I should probably make that more of a priority as that can affect every capture potentially.

Any suggestions on specific avisynth scripts suggestions that would be helpful for ringing-reduction without losing too much detail would be greatly appreciated since I'm definitely not well versed in avisynth as of yet.

Also curious if the extra detail gained by bypassing the noise reduction circuit gives a better starting point for restoration in terms of detail or if it's better to capture the noise reduced output that has less ringing (but also less detail) and go from there?

I still think the DPS and the CVR preNR (pre-noise reduction) looks the most detailed, but they also have the most ringing which U-Matic is quite known for, so I don't think the chain is really adding/creating it, just preserving it as-is.

Just a sanity check...

Sometimes people get overly critical with video quality. (Which is so ironic, because people falsely/wrongly accuse me of that. However, I'm a pragmatist, and a realist.)

What you have to realize is that analog video is give-and-take. Tug on one "error string", and something else unravels.

So, for example, one method resulted in some pretty excessive ringing. So severe, in fact, that Avisynth may not be able to correct it. But the other method results in expected softness, but also some luma adjustments.

If this can be "fixed" by merely using different hardware (and/or better hardware), then that's what you want to do. This is how I'm pragmatic. I use hardware to my advantage.

But eventually hardware ends, and you're forced to software correct beyond it.

Restoring video is about making it better, not making it perfect. Sometimes you have to choose an error as the least worst option. It's too easy to get sucked into a "perfection vortex", which results in nothing ever getting accomplished.

Never use that as an excuse for using crap hardware, or lazy inferior methods. But after you've done your due diligence, tried your best, using the best tools available, you have to stop. Make a choice, convert, and move on.

I see the VH thread, and you're getting help from current/former members of this forum (lollo, latreche34/dellsam34), and a former Doom9 member (Sharc), as I stated. lollo and I disagree on capture cards, but he's quite good at Avisynth, as if Sharc. But neither of them are overly hopeful on Avisynth corrections here.

I think it's your hardware workflow that needs a change. But if you're confident it's the best possible, then you'll need to make a choice, and move on.

I tried using DeFreq in Avisynth http://avisynth.nl/index.php/DeFreq#...erence_removal, but didn't have much luck removing anything. I've used it on diagonal interference before, but maybe the ringing here is too vertical for this tool. The other tool listed in that section of the Avisynth link is FanFilter, and seems to be based on the same principal. Haven't tried that one.

The ringing is kind of like ghosting, which you see on poor over the air recordings. But with U-matic ringing, these ghosts are typically limited to strong vertical edges. There aren't really good tools for ghosting. It's a real tough problem to solve in software. There are multiple Deghosting Virtualdub plugins and Avisynth plugins out there. Again, they can only do so much.

With that said, the one I've had most luck with is LGhost in Avisynth (http://avisynth.nl/index.php/LGhost). What the tool does is overlay shifted copy(s) of the frame at a certain strength in attempt to cancel out the visible ghosting. The side effect, of course, is you may have new ghosts in unwanted areas.

I used the following Avisynth command
LGhost(1,10,20,1,14,20)

In this case, I added 2 sets of ghost "removal" in "edge" mode with an intensity of 20. One 10 pixels to the right, and another 14 pixels to the right. You can disable any pair of ghost values by changing the mode value to 0 (instead of 1 as used in the example above).

I attached a frame capture of before (with the ringing that I focused on most circled in red) and after.

As you can see, it now introduces some ghosting/ringing on the left edge of the frame that wasn't there before.

Feel free to try the other de-ghosting Avisynth filters
http://avisynth.nl/index.php/Externa...#Ghost_Removal

Or you could try the exorcist Virtualdub filter (I think it comes in the Virtualdub 1.9.11 pack that's been posted on this forum before).

I've seen much worse ringing than the bit I see in this clip, if it's any consolation to you. Typically I'm of the mind to capture with more detail, because I know the software denoisers/degrainers will soften. But in this case, if you can find hardware that can reduce the ringing and you want to reduce it as much as possible, I'd say that's your best option.

Tried a couple of random sample uploads to YouTube. It's the PreNR sample that had some of the most ringing. That onscreen text really ends up looking odd no matter what you do, but otherwise I think it looks decent as far as SD NTSC sources from the 1980's go:

https://www.youtube.com/watch?v=RG55gIXXtgw

Would just make sure that 1080p is selected as the resolution for best viewing quality.

This was just a Bob Deinterlace with Handbrake (Mac) and upscale to 1440x1080 along with Unsharp along with some limited cropping that should not affects the original proportions by more than 2%. I set the bitrate to 20Mbps since it's 1080p at 59.94fps. Frames processed per second was around 45, so it's about 50% faster than realtime with a run of the mill M2. You do lose some quality whenever doing this sort of x264 compression, but I think the result looks pretty decent for the relatively uncomplicated Mac-friendly deinterlace.

Video likely could look better with some Avisynth filters like "remove dirt" for the dropouts, but I think this does ok for an "as is" upload route if you are ok with an "as is" upload without editing/restoration.

I did happen to notice one other difference, but I'm not sure if others would really notice without it being pointed out, or at least no one has mentioned it yet.

The DPS unit seems to do a better job as far as a line TBC goes if you advance single frames, but neither appears exactly perfect - it seems that the CVR aligns all the way to the left on every other field, but the alternate fields are off by a little, mainly at the top. If the final deinterlacing output is 29.97 frames per second, this is not visible at all, but if preserving the field rate to 59.97 you can see the flicker of the upper portion of the image on every other frame.

I'll have to try some other TBCs and maybe an ES10 to see if that is just something inherent with Umatic or if other TBCs can align that better on all fields. Also would be interesting to see what the Domesday Duplicator makes of that.

Easiest way to actually see this defect is to look at the youtube link above, go to about the 21-second mark, pause the video and then advance by single frames which you can do with the period and comma keys. The line to focus on would be the vertical line where the camper cab meets the back. this is also the line where you'll see the most "ringing" due to hard contrast between bright and dark in a thin line.

This probably isn't the case, but is there a chance that the deinterlacing process itself is creating that field offset?