Flickering colours on VHS tape?
 

Is there anything I can do to fix the flickering going on in this clip?

An unprocessed sample would be better. This one was converted to RGB at some stage, so the true chroma channels can't be examined.

My thinking would be that heavy temporal noise reduction on the chroma should even out the flickering more, but if it does work you'd have to live with the chroma trailing artifacts that result.

Attached the unprocessed version.

I could be wrong, but it looks like this one went through an RGB intermediate stage too. If you don't mind, please briefly list the workflow used to create this file from VCR all the way to final AVI. Or if it's too much trouble, hopefully someone else has some ideas.

The tape was inserted into a Panasonic NV-HS1000.
The video signal then passed through an S-Video cable to a TBC-1000.
Then through another S-Video cable to a Hauppauge USB-Live2.
Into a PC, where it was captured as a YUY2 Lagarith AVI file in VirtualDub.

What do you mean it looks like it went through an RGB intermediate stage?

I threw VirtualDub's default temporal smoother at it, and it definitely improves things if I crank it up enough. Not sure what 'chroma trailing artefacts' are, so I'm not sure at what point it starts to look bad (I can only play it at one or two frames per second, too, which doesn't help). There is a point at which the outline of the Mitsubishi logo starts showing through the channel 7 logo before the cut, so I've kept it below that line, but other than that...

What msgohan means is that if you "threw" a VirtualDub filter at the it, it performed an RGB conversion (temporal smoother is an RGB filter. It won't work in YUY2). Then the vid was recompressed to YUY2 (?). But that doesn't undo the action of the RGB conversion.

This looks like a multigenerational tape. Whether it is or not, the tape has been abused and/or improperly stored in heated, humid conditions. The "flicker" (which is not really what it is) is a sign that the oxide layer is separating from the adhesive that holds it onto the mylar. The flood of rips and dropouts are one sign of that damage. This could be improper storage or it could be tape that was damaged during play -- literally, it isn't making solid contact against the video heads. I've seen this so many times it's deja vu from so many old damaged family tapes.

You can throw filters at it forever without getting very far, or throw filters at it until you don't have much video loft to watch. You can only do the best you can with temporal smoothers which don't work so well with interlaced video, so use smoothers designed for interlaced source. If you use strong temporal smoothers later in the video when there is motion, you'll soon learn what "chroma trailing artifacts" look like.

Some shortcomings are evident in the capture stage. Colors are clipped above y=16 and the original is seriously over saturated.

Sometimes a video is so badly damaged you just have to do the best you can and live with it. We've all been there. I made a trial run by deinterlacing, shifting chroma around, applying motion smoothers, etc., and reinterlacing. But later parts of the video are bound to suffer from so much filtering. Avisynth and the captured colorspace were required for the attached mp4. Others could get different results and use different methods,

Chroma edges as sharp as what's found in your video are impossible with any consumer analog tape format. They are also probably (definitely?) impossible to send along an S-Video wire even from an ultra-high quality source. Typically these false edges are produced as artifacts from a YUY2 -> RGB -> YUY2 roundtrip, because each conversion step is never perfect.

YUY2 source, UtoY():
Attachment 6590

ConvertToRGB().ConvertToYUY2().UtoY()
Attachment 6591

Unprocessed.avi, UtoY()
Attachment 6592

Maybe instead caused by cranking up sharpness on the Hauppauge proc amp settings? Dunno. But it's somewhere on the digital side of things, not the analog side.

msgohan: some good insight into facets of RGB conversions I've not considered.

How the original sample came to be so saturated or sharpened I can't know. But I doubt the player is doing it on its own. Half the noise level faded when I lowered saturation about 40%. I know that some users pump saturation and.or sharpness to compensate for old and faded tape, but it's usually overdone.

But I'm just guessing. Perhaps koberulz can give more info about this tape's history.

I knew that, I was asking how it could look that way.

Suggestions?

What does this mean?

Trust me, I know I'm not getting this one to look great. I'm largely after 'less distracting'.

I always leave sharpness at default in the proc amp; the only settings I've ever altered are brightness and contrast. The Panasonic has a sharp<>soft slider on the front of the machine, could that be it?

The tape isn't mine, and in fact has existed longer than I have, so I can't help you with its history.

The original "flickering" looked like chroma noise from reusing a tape, where the recorder lacked flying erase heads.

The next "unprocessed", however, show it to be just plain chroma noise. Nothing special, nothing to do with reuse.

This is why it's important to always give unprocessed video, when showing samples. Sometimes attempts to fix it can muddy the situation. And in this case, make repair harder.

- The 1st issue would have been hard, while the 2nd really is not.
- The 1st issue is repaired best by temporal chroma NR in Avisynth and/or VirtualDub, while the 2nd is best fixed with VirtualDub CCD.

There's also a lot of chroma offset, and Avisynth is needed for that. (Best method, at least. CCD can somewhat hide it, but not entirely.)

Color being clipped or not clipped is probably the source or the tape, not the conversion method. Or both. Or all of the above. VHS is crummy. It can be made better, of course. But I use my eyes, not meters and graphs. In this short sample, I'm not willing to adjust colors. It's fine.

I don't have an issue with the chroma edges. I don't see halos.

So what am I doing that's making it worse? I've already got CCD in the VirtualDub chain that produced the first sample. It definitely looks worse than with VDub's default temporal smoother added (which gets me in the ballpark of sanlyn's sample).

I've used flaXen's VHS filter in VirtualDub to handle the chroma shift. What's the reason for preferring AviSynth?

Not sure.

I can't open your files in VirtualDub. Is that Lagarith, and not Huffyuv? On my current video system, to appease the codec gods, I've not installed anything that I don't use. I'd need a Huffyuv conversion.

I can't open HuffYUV in Premiere, so I always use Lagarith.

I have no issues with Premiere. The fix to install both x86 and x64 together was discussed in another thread.
See here: http://www.digitalfaq.com/forum/vide...html#post44169

This was an easy problem. :)

See attached. I could have tweaked it more, but this is mostly just proof of concept.
- Avisynth offset fix (Flaxen is lousy)
- Avisynth CNR
- VirtualDub CCD
- VirtualDub Chroma NR
- VirtualDub CCD again

I also added a stabmod() and my script to remove most tracking/dropout issues. This last one is a bit hard to explain, not ready to do it yet. Even on Skylake CPU, these drag it down to 4fps, so not for weak systems!

See clip attached. MP4 @ 15mbps, AVI not needed, huge.

You're right, CCD was not enough. So it may still be an erase head issue.

Not really sure I follow that. Is there any reason to go with HuffYUV over Lagarith?

What's the actual code in AviSynth to use those? I assume 'CNR' and 'Chroma NR' are both short for 'chroma noise reduction'?

What are you achieving by running CCD either side of Chroma NR in VDub?

What's lousy about Flaxen?

What does stabmod do? I Googled it and found your AviSynth guide thread in the General Discussion section, but there doesn't seem to be much info on what it actually does.

Very nice work LS, but the chroma does trail into the shot change over 4 frames before disappearing on the 5th. We need a sample from koberulez with real motion to see how much artifact balancing is truly required.

No. As I mentioned, it's happening after the analog signal has entered the digital domain: at the Hauppauge digitization step or thereafter -- leaning heavily towards "thereafter", especially now that you've confirmed Hauppauge sharpness and saturation were at default. Even if something like the TBC-1000 was converting to RGB internally, or you cranked sharpness to insane levels with 10 video processors chained together, the chroma edges would still end up softened before and/or at the capture stage because of the limitations of S-Video and capture device samplers.

I don't know whether this makes things any more clear, but the first test pattern in each "VHPatterns2" column in this thread includes a patch for checking chroma resolution (on the bottom, particularly the one labelled "1.5"). If you look at the greyscale chroma view, you can see how blurry even the best capture devices that I've used are, as compared with the super-sharp chroma edges found on your capture. (I discovered that the DVD player I used is partly at fault for reducing the chroma res, but that's another story: most of the capture devices can't even keep up with the reduced level put out by this player, and DVD has much higher chroma bandwidth than VHS. VHS captures "should" always have blurry chroma.)

There are luma halos. Most visible along the "P" in "Perth", I think.

These added chroma edges from roundtrip conversion are invisible when actually watching content. The only reason I brought it up was because the added conversions hinder analysis and possibly processing. Had you guys responded with filtering suggestions first I'd have kept my mouth shut and left it to the restoration experts. :P

Thanks for the links to that chroma testing. If you're curious about the mp4 I posted it was done entirely in YUV, mostly with a few passes of dfttest and FluxSmoothT -- not that I'm happy with the results. I treated it as both progressive and as interlaced and got exactly the same results, as it seemed originally telecined and incorrectly deinterlaced, then the VCR played it as interlace while mediaInfo sez progressive.

That video has certainly been around a few neighborhoods, at least one of which was digital, and looks as if it went from tape to digital and back to tape, then maybe sat in the sun on someone's window sill or in an auto's trunk for a few summers. I echo the requests, what else is on this tape? Is it a TV show? Are there people, animals, cars, motion, what?

Not sure how helpful a motion sample will be; quality is all over the place between the opening graphics package, the in-studio introduction, and then even each individual camera angle through the game. I was fiddling with the chroma shift earlier to get the Mitsubishi logo right, and now it's completely out of whack during the main program.

So this one's a serious case of multiple combined restoration efforts being required, and as such posting a clip from the game itself probably isn't helpful with resolving a graphics issue. All the motion there, and the general lack of any graphics at all, should mask the chroma noise a bit anyway meaning lower settings can be got away with in order to preserve everything else.

In any event, I can't put together samples right at this moment.

EDIT: sanlyn posted while I was typing. It's a basketball game from 1985.

To be honest I didn't check it out until now, but it looks like a nice result as well. Since the sample doesn't really feature motion I wasn't too interested in checking out the restorations. :o I only bothered downloading Smurf's because he mentioned fixing tracking/dropouts. (I really am that lazy.)

lordsmurf, I'm trying to download your AviSynth filters pack from the guide thread, but it's failed twice. Is there an issue with it, or is the problem at my end?

Avisynth:
VirtualDub .vcf attached.

The dropout/tracking fix must wait until later. It's slow and unstable. If you have the newest Skylake CPU, I can look to add it. If not, don't bother, it will crash.

This is clearly some sort of nth gen madness going on. I'd need to see the whole tape to know what's going on. It cold be reuse, bad analog editing, etc. There's to many variables without physically seeing the tape, or at least the full video. And no, I really don't want to see the whole video, please don't try to upload it.

Post the link of the download page where you're trying to download the filter pack. ;)

BTW, my video system is busy right now, can't really do anything else. I did the above test clip between tasks.

CNR is the name of an Avisynth filter.

@msgohan: The chroma leaking across frames is there, yes, slightly. Like I said, proof of concept, it still needs some tweaking. Having 2-3 frames of chroma blend is going to be the best you can do here. CNR was set to wide mode, and that is probably why.

I've never even heard of a Skylake CPU.

By way of providing data points: the reason I capped on my Panasonic rather than my Philips is I had really bad tearing on the latter. The chroma shift was also significantly worse. What would physically seeing the tape tell you?

Filter pack is attached to the last post here: http://www.digitalfaq.com/forum/news...-avisynth.html

I just tested the link and it worked.
Here is a copy of the download link (left-click on the link): http://www.digitalfaq.com/forum/atta...isynth-plugins

Download the RAR to a separate folder, not to the Avisynth plugin folder.

If you get an error message, what does it say?

It was just failing in the web browser. That link worked though.

lordsmurf, could you clarify the HuffYUV vs Lagarith and HuffYUV installation thing?

I use Lagarith for many of my working files saved as YV12, which huffyuv doesn't support. I don't use Premiere. But I've had no problem with huffyuv in Adobe AfterEffects. That's Adobe for you.

New super CPU. But I'm tired of building new PC's that are obsolete as soon as they're set up and working.

Other than skewing erors, you should see just about the same fluttery disturbance and saturation you see with the Panny in this case.

VDub gives an error regarding Camcorder Color Denoiser 1.7 when I try to load the VCF file: no such filter loaded. The one in my filter chain is a Camcorder color denoise 1.6. Am I using an older version of the same filter, or are they different things?

newest ccd here (left click): http://www.infognition.com/cgi/getfilter?id=220
Not very effective in this case, however. Best to work in the original YUV before trying RGB.

Many more VDub filters here: http://www.infognition.com/VirtualDubFilters/
But watch out for the VirtualDub Filter Pack on this page. Many in the pack aren't updated.

lordsmurf, I assume CNR is CNR2? You didn't give that in your AVS script, but you mentioned it when describing your workflow.

Attached HuffYUV samples from the pre-game intro video, the game itself, and a studio segment.

EDIT: Installed that plugin, still getting the error.

EDIT 2: It helps to put things in the right folder. Can now load the VCF.

EDIT 3: Can't see the issue pointed out by msgohan, but on the Intro clip the red jacket on the guy who does a layup just before the titles come up sticks around way too long. Unchecking 'wide' under V on CNR in VirtualDub helps but doesn't completely eliminate the issue, removing CNR from the filter chain completely does.

The ccd I had to use for lorsdsmurf's .vcf file was downloaded from the link I posted. Iyt's not the version I normally us, so I saved my old one. I normally use CCD_SSE2.vdf listed as "Camcorder color denoise 1.6 MT". The version downloaded is ccd.vcf dated 5 September 2013.

Thanks for the samples. Apparanetly processed previously by a couple of dudes who spent days in a closed room with their hookah pipes loaded with crystal meth, conducting their own video Demolition Derby. What a shame. Well, we can give a try later. But the vids are pretty well borked. AMybe the default settings for your capture device aren't optimum, but I wouldn't know what to set for. Try saturation for beginners.

I was a little too quick on my conclusions about that red jacket; it was visible a little even in the AVS file. So I dropped un and vn in CNR2 to 35 from the default 47, and it disappeared. Re-enabling CNR in VDub brings it back though. I'm not sure how to fiddle with those settings, other than checking and unchecking 'wide'. There's an explanation of the X and Y axes in the settings for it, but I don't understand what they mean.

What leads you to the conclusion that it's been processed?

What you can do is load the other filter (Chroma Noise Reduction) by itself using the revised .vcf file attached below. Then load two copies of your own CCD filter. Set one CCD filter for a strength of 50, set the other for a strength of 30. You can chain multiple copies of the same VDUb filters together.

A .vcf file is a plain text file. You can open it with Notepad to see what it's doing. You can make changes, but NEVER overwrite the original. Save the changes as a new .vcf with a different name.

All you have to do is watch it. But don't ask what was done to it. It's either been wildly processed or was recorded with a VCR that needs critical therapy. The tape has also been improperly stored for quite a spell. There are color stains all over the place from what appears to be storage in humid or heated conditions. It looks cooked.

Indeed. It does prevent hassles, LOL! I've done it myself. BTW, you can have both versions of the CCD vdf installed if they have different names.

Did you miss my earlier edit? I was putting the plugin into the wrong folder because sometimes I'm an idiot.

What would capturing with lower saturation do that lowering saturation in software won't? What's actually wrong with the saturation?

Yes. I changed my last post as well. Fast and furious today, what?

Either way. With advanced color controls later on you can adjust saturation for individual colors, but not during capture. Better to try it YUV fist, though, if possible.

A bad VCR and unsuitable storage conditions seem more likely than processing.

I mean, yes, it looks awful. I know enough to be able to tell that (as do people with far less knowledge and experience than me, probably). It's the conclusion that it was processed that I don't really understand. I just looked at it and figured that's what a crappy 80s tape looks like. Apart from the horribly red ground-level cameras, which I'm guessing is an issue with the camera settings and nothing to do with the tape at all. Going through in Premiere and cutting it up to use a different AVI for each angle has been tremendous fun, particularly when they add replay wipes. *sigh*

What do you mean by that? Processing through AviSynth before VDub converts to RGB?

Skylake aka i7-6700K: https://www.amazon.com/Intel-Unlocke.../dp/B012M8LXQW
Also: http://www.newegg.com/Product/Produc...82E16819117559

Still the current CPU, 1 year later! It really is that good. :)

I see the Avisynth post: http://www.digitalfaq.com/forum/news...html#post43574

The name was "lordsmurf's Avisynth". But the forum broke the name. Not sure if it's also screwing up the downloading. So I just deleted it, re-added it as "lordsmurf Avisynth". I've probably added some stuff in the last 6 months. Need to give that thread attention again, new beta download. BTW, that Avisynth guide is being co-developed with the new glossary, as there's much overlap.

I always start with my current multiscript, and delete unneeded lines to post it. I must have deleted the CNR2 line?
CCD 1.6 and 1.7 do the same thing. No real discernible difference that I can remember off-hand.

Wide for VirtualDub CNR increases the temporal axis. (I don't use it much, so want to double-check that statement.) Removing it reduces both negative and positive effects. You can play with other value, but I find that the defaults are often best. If those don't do what's needed, no other settings will either.

I'm not at my video system now.

Sorry for the delay. Medical stuff today. Phooey.

Yes. Usually do as much as ia pracrticable in the original colorspace before making colorspace moves.

I made a trial run with the studio avi to see what could be salvaged from the wreckage. Attached as "A" below. A test, really, because I have no idea what to do with all the pesky horizontal dropouts and comets without destroying what little remains. It's a hassle to keep the scant detail level of the tape from looking like plastic. I lowered midtones and darks to get a little depth into the image, but this might not work if there's dark sports playback during all the talking.

Attachment "B" is my feeble attempt at clearing the dropouts with RemoveSpotsMC3. There's some detail loss but IMO you don't get that much back in return. Now need a brilliant idea for those dropouts. And why is that emblem in the upper right so much noisier than the rest of the frame?

I had to use two steps (two separate scripts and two intermediate files) to get the "A" results. I can clean up the messy scripts and post later if you want.

This tape isn't really a challenge.
The colors are illegal still, as I only addressed the noise:

Attachment 6605

Vdub crashed 3 times out of 4, to run my script. But it did run. That script fixed most dropout and momentary tracking problems. We can discuss it later this year.

The chroma is already destroyed. There really are no fine details in the source. What you're seeing is noise. When you remove it, you think you're also removing detail. But you're not. I always try to discern noise from detail. Sometimes it's not easy, and we all make that mistake.

Thanks for the notes.

No, there's no detail. So I added noise (AddGrainC).

The red jacket in the image tops out at RGB 244, but is mostly around RGB 220-230. The white shirt hits 244. The jacket is still over saturated and looks neon. Unfortunately if red is curt any further the guy's face turns green, then cyan In short, the core image chroma values were wrecked a long time ago. RED is clipped in YUV. More tweaking in RGB obviously due, but you can only go so far with thin data and clipped colors.

Are those image from the original? The mp4's have a red jacket under RGB 200 and the shirt collar is white, not cyan.

But I demand my free help immediately!

The only issue with that being I know nothing about color correction in AviSynth. I can handle most things okay(ish) in VDub, so that's where I've preferred to work. Unfortunately AviSynth isn't really something you can just play around with; the ability to drag sliders around and watch a video preview change in real time makes it much easier to learn things.

Even the AviSynth wiki - which still requires you to actually know what you're looking for to a large degree - is pretty light on explanations for the most part.

Well, I'm figuring in any case that the studio segment and the game will each need their own settings. Not sure if the intro will fit into one or the other or require a third (or perhaps just both; masked so that one affects the video area in the middle and the other affects the graphics around it).

I'm putting together an MP4 of the entire opening sequence, just to give you an idea of how it all works. I'm basically just grabbing the first and last few seconds of each segment, but to catch everything and show the sequence it still needs to be way too long to post as an AVI. There's nothing in it that's not in the already-posted AVI files though.

Be great if you could.

I still don't really get this 'oversaturated', 'illegal colours' thing. I can see, obviously, that sanlyn's sample is less saturated. And yes, the jacket looks less ridiculous. But you're both talking like there's objective realities about saturation where I'm only aware of it being a matter of taste. I was walked through histograms and wave form monitors for luma, so I get that...obviously there's a chroma equivalent?

I did some quick jobs based on MPG files a while back, and I've noticed that I tended to have the brightness too low and the contrast too high, because I was going in with a nooby 'make the blacks black and the whites white' mindset (and didn't have the color correction or noise removal skills to make almost-white areas less oddly-colored without just blowing them out). I had the opportunity to work with one of those in combination with a DVD produced from the original broadcast tapes, and noticed my versions were also significantly more saturated, so presumably there's a similar issue happening in that respect.


What does this actually mean, and how do you determine it? Pulling screenshots into Photoshop, or is there a quicker way? AvsPmod gives hex values on mouseover, but that seems less helpful. EDIT: Okay, turns out you can change that in settings, so I switched it out for both YUV and RGB. However, that's not a solution once we get into VDub.

*blank look*

Unrelated edit: For some reason the timezone settings don't seem to work. Apparently I made this post at 8:22pm yesterday, when it was in fact 9:22am. Today, obviously. I have it set correctly in my control panel.

MP4 attached.

I know you're joking, but we do get that mentality around here almost monthly. :screwy:

In the last 30 days:
- Free Member, daily replies, "please I need help now!"
- email, (not) potential customer, "I need a phone conversation, and you need to answer these (25+) questions" ... for a $25 project
- PM, "can you restore this MP4?", sure, that would be about $65, "I don't want to pay for it!!!!"

Some people are f'ing nuts. :screwy:

I seriously cannot click this enough: :screwy: :screwy: :screwy: :screwy: :screwy: :screwy:

I hate cc in Avisynth, and avoid it. VirtualDub is okay, but true correction is reserved for Premiere. I'm trying to learn the free version of Blackmagic Davinci Resolve.

Use AvsPmod.

Yeah, it's really crappy. This has been my main complaint for 15 years now. I used to get in heated discussions with plugins devs at VH, because their documentation was obtuse crap -- or simply missing entirely. Making a plugin is great, but worthless if nobody can use it. Several filters are still total unknowns after a decade. Nobody knows to use it. Waste of space, really. Just a tease.

Some are actually just "work in progress" with no progress.

I plan to address that with my Avisynth documentation. Hopefully I can tap others, like sanlyn and msgohan, to help on it.

Welcome to complex scene-based restoration. :laugh:

Colors should not be neon. The saturation is "illegal" because it blooms, destroys detail. For example, the "studio" lapel. You can see them. All you can see is neon pink. The value is exceeding the luma that held the primary contrast data. It's not legal, not balanced.

Not in this instance. Again, it chroma blooms beyond the luma. That should not happen.

Pay close attention to the VCR, TBC and capture card for harsh value changes. Process of elimination needed to exclude items. Yes, that means multiple equipment. And I know, not possible for everybody. You just need to extra vigilant in not screwing up values.

sanlyn and I said the same thing, but with different words. He likes jargon, which is good. That lets me go more jargon-less. So do you understand it better now, with my explanation? If so, use his to better understand the jargon, which you'll see in the software.

The forum may be hard-coded to CST. The vB clock is temperamental. :hmm:

My settings are actually conservative, and would work for that entire tape. Fix noise now. Do color correction in a second pass. It's lossless, after all -- and that's why.