VHS videos appear to be damaged - can only deinterlace with blend

[Ferrari420]

I have two tapes saved to my PC, three actually but can only upload two samples and all of them show patterns if you try to deinterlace them with any method other than blend. For the purpose of this post, I'd like to deinterlace them to 25fps as that is what my family prefers. 50fps appears too fast compared to the playback on the VHS tapes. Last time I did this project, they commented that playback was too fast. 25fps is the closest to the tapes.

If they can't be deinterlaced without showing those patterns, will the quality be better if it is left interlaced, even if it is not deinterlaced (because deinterlacing even with hardware deinterlacing causes artifacts)? Is interlaced preferable to blend? I will be playing back on DVD but the whole point of deinterlacing is to eliminate the lag hardware deinterlacing creates, which my family also commented on. Every time they pause and play, there is a delay which they find annoying.

Are the tapes damaged? Both fields - top field and bottom field show strange colours.

[Ferrari420]

Sample video 3 attached

[Sharc]

There are no blends. The rainbows and dotcrawl may have been caused by the composite signal with luma<->chroma interference. Did you switch the CombFilter ON of your DVD recorder?
Some quick improvement here. Interlaced 25fps.
Interlaced vs deinterlaced has been commented enough in your other thread. No need to repeat everything here.

FWIW here the script:

AVISource("Sample Video 1.avi")
AssumeTFF()
converttoYV12(interlaced=true)
separatefields().vinverse2().weave()
QTGMC(preset="slow")
Derainbow()
MCDegrainSharp()
QTGMC(InputType=2)
Separatefields().selectevery(4,0,3).weave()
converttoYV12(interlaced=true)

[Selur]

I agree chroma is messed up, there's what happens if you throw Vinverse2, QTGMC,Cnr2, BasicVSR++ (with and edgeMask) and SMDegrain at it,... (SMDeGrain is only there because I forgot to disable it) Script: https://pastebin.com/NSnMuJEg
Color wise Vinverse2 and Cnr2 do the work.

Cu Selur

[lollo]

Interlaced vs deinterlaced has been commented enough in your other thread. No need to repeat everything here.

Well, there is something really bad left in that thread needing a reply, for other readers

https://forum.videohelp.com/threads/412746-What-s-the-best-deinterlacing-method-for-VH...=1#post2718412

The 2 fields may however be pictures from the same instant in time, hence there is no motion between the 2 fields of a frame.

Preach, preach! This is why the dogma of "deinterlacing to 25/29.97 is throwing away data" is nonsense.

In an ideal world, yes, you'd always deinterlace to full temporal depth. Of course that requires interpolation vertically, and creating new data can be just as bad as "throwing away" data. But the dogma is just BS, because it's not an ideal world. There are specs, and there are artifacts due to interpolation. You do as best as possible, meaning you sometimes deinterlace to the source frame rate. This is why QTGMC has many settings and options.

lordsmurf, you must be joking:

1- when the fields are at the same moment in time ("progressive" video), there is no deinterlacing required. So the processed video after filtering will stay at its original frame rate. The only exception is when you have a mix of interlaced/"progressive" and you decide to deinterlace everything to go fast.

2- when the fields are at different moment in time ("true" interlaced video), the deinterlacing need be at double frame rate, otherwise you lose (almost) 50% of the temporal data. Almost because there is the Kell factor to consider.

I explained that to you many times.

What is really a BS, actually, is not the dogma you mentioned.

[mammo1789]

Virtualdub Camcorder color denoise will do much better job than CNR or Viniverse on the chroma noise I am out for holidays so cannot test it but I am sure that it will remove the noise with little damage

[Selur]

CCD (Vapoursynth version) doesn't really help with that sample here, but maybe the VirtualDub version does.

[Sharc]

Virtualdub Camcorder color denoise will do much better job than CNR or Viniverse on the chroma noise I am out for holidays so cannot test it but I am sure that it will remove the noise with little damage

Much better? Hmmm, I tried it as well. Horizontal stripes, rainbows and dotcrawl still remain.
Less damage? It's an RGB filter which converts YUV to RGB (and back for encoding), which inherently introduces some losses (which I agree are usually meaningless for VHS, unless the conversion introduces clipping ....).
So looking forward to seeing your example when you are back

[Bwaak]

when the fields are at the same moment in time ("progressive" video), there is no deinterlacing required.

Deinterlacing is always required to display interlaced video on a progscan monitor. Your "no deinterlacing" is actually "weave" deinterlacing, the standard method of Youtube and many software video players.

[lollo]

when the fields are at the same moment in time ("progressive" video), there is no deinterlacing required.

Deinterlacing is always required to display interlaced video on a progscan monitor. Your "no deinterlacing" is actually "weave" deinterlacing, the standard method of Youtube and many software video players.

My "no deinterlacing" refers to the post-processing flow for filtering, that is what we were talking about, dear professor

[Bwaak]

when the fields are at the same moment in time ("progressive" video), there is no deinterlacing required.

Deinterlacing is always required to display interlaced video on a progscan monitor. Your "no deinterlacing" is actually "weave" deinterlacing, the standard method of Youtube and many software video players.

My "no deinterlacing" refers to the post-processing flow for filtering, that is what we were talking about, dear professor

I don't know who are those "we", but unless you are engaged in a private conversation you may want to make your messages less ambiguous.

If, when loading an interlaced video and without explicitly deinterlacing it, the resulting file is marked as progressive, this means that a default deinterlacing method (usually "weave") has been applied behind the scenes, and it is a bug, and it confuses newbies.

[lollo]

I don't know who are those "we", but unless you are engaged in a private conversation you may want to make your messages less ambiguous.

My messages are ok.

If, when loading an interlaced video and without explicitly deinterlacing it, the resulting file is marked as progressive,

loading to what?, marked by whom?, ...

and it confuses newbies.

The only newbie here is you.

We (Sharc, Skiller and myself, so you can understand) were talking about processing an interlaced file where the fields are coming from the same moment in time.

Next time I will draw a picture, especially for you.

[Bwaak]

If, when loading an interlaced video and without explicitly deinterlacing it, the resulting file is marked as progressive,

loading to what?, marked by whom?, ...

To whatever and by whatever software you use.

We (Sharc, Skiller and myself

Thank you for the clarification.

Next time I will draw a picture

If you cannot make it clear using words, definitely draw a picture. It may or may not help, depending on your drawing skills.

[lollo]

If you cannot make it clear using words, definitely draw a picture. It may or may not help, depending on your drawing skills.

The only thing in doubt here is your intelligence, not my skills!

[Bwaak]

If you cannot make it clear using words, definitely draw a picture. It may or may not help, depending on your drawing skills.

The only thing in doubt here is your intelligence, not my skills!

Who's questioning my intelligence? Someone who needs to illustrate his words with pictures?

I guess we can do this all day

[lollo]

Who's questioning my intelligence? Someone who needs to illustrate his words with pictures?

Yes, because your culture is very limited. In Italy (and somewhere in Europe), when somebody does not understand a simple thing, we day "I will then make you a picture". It is just another way to say "you're so stupid..."

I guess we can do this all day

Yes. I will teach you something else. In Italy we say: "never argue with an idiot, he drags you down to his level and beats you with experience".
Now that you learned something, I will stop here, to remain at my level!

[Alwyn]

50fps appears too fast compared to the playback on the VHS tapes. Last time I did this project, they commented that playback was too fast. 25fps is the closest to the tapes.

I'm curious about this. My understanding (and experience) is that a 25fps file will play at the same "speed" as a 50fps file; the advantage of the 50fps file being smoother motion because you have double the number of frames to display the motion. Are you doing anything more than simply using the deinterlacing filter?

Re the colours, it is quite possible that the tapes were recorded like that; I have a dozen tapes that I re-recorded years ago with a rogue VCR with bad banding which only came to light when I tried to deinterlace them.

FYI, here's what the Neat Video denoiser (Paid) can do to your sample Video #2 with a few clicks:


[Attachment 76004 - Click to enlarge]

Well, there is something really bad left in that thread needing a reply, for other readers

lordsmurf, you must be joking:

@Lollo, please post this over on the other thread then. Don't pollute yet another topic with your war against LS.

@Lollo and Bwaak, please take your bickering somewhere else. You're ruining another good topic with your inane arguing.

[davexnet]

This post mentions Temporalsoften, seems to help the chroma issues in Sample Video 2.avi from post #1
https://forum.doom9.org/showthread.php?p=398603#post398603

LWLibAvVideosource("c:\users\davex\downloads\Sample Video 2.avi")
bwdif(field=3)
TemporalSoften(2,0,255)

[lordsmurf]

edit

Just saw this:

@Lollo, please post this over on the other thread then. Don't pollute yet another topic with your war against LS.
@Lollo and Bwaak, please take your bickering somewhere else. You're ruining another good topic with your inane arguing.

Yes, agree, let locked conversation stay locked.

/end

____________________________________

lordsmurf, you must be joking:

No, specs and artifacts can be the limiter. The "throwing away" dogma (videophile type BS) is stupid, and we have to make the best of it. I can only assume that I've come across more/varied VHS tapes than you have, and dealt with more output needs.

1- when the fields are at the same moment in time ("progressive" video), there is no deinterlacing required.
2- when the fields are at different moment in time ("true" interlaced video), the deinterlacing need be at double frame rate, otherwise you lose (almost) 50% of the temporal data. Almost because there is the Kell factor to consider.

We must not be referring to the same thing. You refer to truly progressive footage encoded interlace. I'm not. I refer to interlaced video that contains non-moving points in time. From a technical/dogma/nerd stance, yes, the data is discarded. From a visual "using your own eyes, using your common sense" approach, nothing is lost.

Again, read what I wrote:

In an ideal world, yes, you'd always deinterlace to full temporal depth. Of course that requires interpolation vertically, and creating new data can be just as bad as "throwing away" data. But the dogma is just BS, because it's not an ideal world. There are specs, and there are artifacts due to interpolation. You do as best as possible, meaning you sometimes deinterlace to the source frame rate. This is why QTGMC has many settings and options.

We don't disagree, but I just push back on the dogmatic nonsense (not even necessarily from you, just "out there"). Above all, I'm a pragmatist. I seek quality, but without being dumb about it. Nor do I let laziness/cheapness derail the goal (which is equally dumb as dogma). There must be a sensible middle. This is it.

[Ferrari420]

If you play the samples as is that is what they look like on the tapes. The goal is to make the samples deinterlaced without the patterns appearing. Blend achieves that, but why? Blend combines two fields into the same frame, right? The patterns appear only when deinterlacing with any other method. Now, why do they appear even when not discarding the other field, say by doubling the frame rate with Yadif? Maybe combining deinterlacing methods can fix the issue, or some kind of filter? When the two fields are combined it forms a consistent colour. When one field is removed it leaves a pattern. I've never seen this before.

I know that it can be fixed with AviSynth, but I am not ready to start learning scripting yet.

On a side note could we please stay directly on topic?

[Selur]

when using ffmpeg, something like 'hqdn3d=5:4:7:6' should also help with the flickering.

[Sharc]

The goal is to make the samples deinterlaced.

Goal? Not recommended to deinterlace your interlaced video. For DVD leave it interlaced, 25i, means 25 interlaced frames per second= 50 fields per second, perfectly DVD compliant. But we have been through this in your former thread. For some obscure reasons like "plays back too fast" you stick to deinterlacing to 25p it seems. So be it.

Blend achieves that, but why? Blend combines two fields into the same frame, right?

What "blend" are you talking about? Which filter(s) did you use?
Combining the 2 fields into a frame is called "weaving". On a progressive monitor you see the typical combing, unless it gets deinterlaced on the fly (real time) at playback time by the player/TV - which is what happens for DVDs. Player/TV deinterlacers are at least as good as yadif.

The patterns appear only when deinterlacing with any other method. Now, why do they appear even when not discarding the other field, say by doubling the frame rate with Yadif? Maybe combining deinterlacing methods can fix the issue, or some kind of filter?

Both fields are unfortunatley 'damaged' (rainbows and dotcrawl). There is no clean field.
Step through the attached video and you will see that both fields are contaminated.

I know that it can be fixed with AviSynth, but I am not ready to start learning scripting yet.

Not really fixed without a compromise, but mitigated, yes. Learning Avisinth would be a good investment of your time. Until then someone else may chime in with some advice for a GUI.

[lollo]

@Lollo, please post this over on the other thread then. Don't pollute yet another topic with your war against LS.

Did you ever notice that that thread has been closed?
BTW, I am not doing any war with my friend lordsmurf, we simply disagree on some topics and write our considerations. I think this is a constructive discussion, allowing people to build their own conclusions.

@Lollo and Bwaak, please take your bickering somewhere else. You're ruining another good topic with your inane arguing.

I already wrote that I would have stopped there.
What is really insane in this topic is your "yet another" attempt to promote a commercial software used with few clicks (and not at its full potential) for a minimalist approach.

No, specs and artifacts can be the limiter. The "throwing away" dogma (videophile type BS) is stupid, and we have to make the best of it.

The "throwing away" is science, as you will see later.

I can only assume that I've come across more/varied VHS tapes than you have, and dealt with more output needs.

Your assumption, which is also wrong. See my example later (Based on the facts, I could say to you "I have dealt with more/varied VHS tapes than you have", but I won't, because you're my friend )

Again, read what I wrote:

In an ideal world, yes, you'd always deinterlace to full temporal depth. Of course that requires interpolation vertically, and creating new data can be just as bad as "throwing away" data. But the dogma is just BS, because it's not an ideal world. There are specs, and there are artifacts due to interpolation. You do as best as possible, meaning you sometimes deinterlace to the source frame rate. This is why QTGMC has many settings and options.

We don't disagree, but I just push back on the dogmatic nonsense (not even necessarily from you, just "out there"). Above all, I'm a pragmatist. I seek quality, but without being dumb about it. Nor do I let laziness/cheapness derail the goal (which is equally dumb as dogma). There must be a sensible middle. This is it.

In the real world what you claim does not exist. A video always contain some motion. If by any chance there is no motion in time (maybe filming a stone in a valley), the 2 fields show the same content even if they are interlaced, and that's obvious.
If you discharge one of them there in nothing to loose. But in that case you do not even need to deinterlace, so the whole argument is useless. (BTW, deinterlacing to "interpolate" and discarging a field/frame is lower quality than leaving everithing as it is, so you are wrong even here).

And now an example, instead of your common blah-blah without facts. This is an interview to some actors, recorded in SVHS (not by me) and captured (not by me), that I am processing for distribution.
The major part of the scene is almost fixed (very few object move) but there is some motion. And here a comparison between a processing with QTGMC at double frame rate and QTGCM discharging a frame.
The result is obvious.
P.S. in the comparison the single frame rate has been duplicated to match the double frame rate sequence.

comparison (look the hand moving):


original captured video: SuperVHS Amarec + Panasonic.avi

double frame rate processing: dfr.avi

single frame rate processing: sfr.avi

And finally, I understand that we agree more than appears in our posts and that some people are not interested to obtain the maximum quality, and are somehow satisfied even with inferior procedures, but I will always promote the best (to my humble understanding) approch.

[lordsmurf]

And finally, I understand that we agree more than appears in our posts and that some people are not interested to obtain the maximum quality, and are somehow satisfied even with inferior procedures, but I will always promote the best (to my humble understanding) approch

Again, sometimes specs and artifacts tie our hands. Dogma only serves as a tool to chastise in those cases, often from a peanut gallery that fails to grasp the problem. I've actually discussed some in minute detail here before, but I don't have the time to look up my past posts from years ago. In general, yes, we agree, full temporal depth is ideal (sort of how 720x is best, even if source is less than 352x; no reason to compress for file size in the 2020s). I refer to edge cases here, no real choice, either accept the not-bad (but dogmatically "wrong") video, or have problems that require excessive work, or are impossible to resolve.

But in that case you do not even need to deinterlace, .

This is starting to get pedantic. I decline. Let's address the OP's issue instead.

[lollo]

I refer to edge cases here, no real choice, either accept the not-bad (but dogmatically "wrong") video, or have problems that require excessive work, or are impossible to resolve.

Nobody here applies "dogma" without testing. And if you remember my posts, I always recommended to everyone to do his own extensive testing and never blindly trust anyone

Let's address the OP's issue instead.

Sharc is doing an excellent job supporting the OP in all the required aspects, there is no need to add anything

[Alwyn]

Did you ever notice that that thread has been closed?

Irrelevant (although not unexpected given the usual suspects melted down yet again). If you couldn't post on that thread, start another to continue your battle, don't hijack this one.

What is really insane in this topic is your "yet another" attempt to promote a commercial software used with few clicks (and not at its full potential) for a minimalist approach.

Point me to the statement that any posts/suggestions on VH must only include the use of free software.

I'm pleased to read that I could do even better with Neat if I used it's full potential. I have work to do. Actually, I don't. It works great for me, it is not a minimalist approach, and it is a pity that you are so pigheaded that you cannot accept that others might not need or want to achieve what you might be able to achieve with your expert knowledge of AVISynth. Clearly, you have no understanding of the concept of cost-benefit analysis. Many, many hours trying to come to grips with AVISynth and more specifically the hundreds of filters and combinations, verses spend less than $100 and get the denoising job done quickly and easily and move on with life.

I've said it before and I'll say it again; I am severely impressed by what you guys can do with AVISynth. It is a pity that you can't accept that some people may not want to (or be able to) do what you do with it.

[lollo]

I've said it before and I'll say it again; I am severely impressed by what you guys can do with AVISynth. It is a pity that you can't accept that some people may not want to (or be able to) do what you do with it.

Alwin, I already wrote you once that your approach is fine, for you and for many others, and I understand your point of view.
My words were just a (probably bad) reaction from my side to your criticism about the words I had with the "professor".

edit:

If you couldn't post on that thread, start another to continue your battle, don't hijack this one.

I did not hijack this thread, there was a reference made by deinterlacing aspects being covered (which was not) and I could not write there because the thread was closed.

[Bwaak]

If by any chance there is no motion in time (maybe filming a stone in a valley), the 2 fields show the same content even if they are interlaced, and that's obvious. If you discharge one of them there in nothing to loose.

Depends. If the video has been acquired with a video camera having a native interlaced pickup, then you are right, as both fields would be the same. But if the video has been acquired with a video camera having a full-frame native progressive sensor, or from film, then by throwing away one field you would lose vertical resolution.

But in that case you do not even need to deinterlace, so the whole argument is useless.

As I said before, you always need to deinterlace if your target device is progscan display.

If your software generates progscan video from interlaced video without you explicitly instructing it to deinterlace, it is a nasty side effect that I consider a bug.

Back when I used CoreAVC, it has clearly listed all the deinterlacing options, including None (Weave). Since the target display was a computer screen, you had to choose one of the deinterlacing options. The only flat-screen technology that could natively displayed interlaced video was Hitachi ALiS.

Whenever I use video captured with VirtualDub2 on a Vegas timeline, Vegas reports it as progressive with PAR 1:1 no matter the codec, I suppose because AVI does not have a well-defined way to describe this in the metadata, or maybe this is just a VDub's bug. So I have to remember that it is interlaced PAR 10/11, or just visually check each clip, which is a nuisance.

[Sharc]

Guys, the OP's (Ferrari420) captured video is truly interlaced - as it should be, full stop.
The OP should understand the consequences when deinterlacing it to 25p, and he should understand that 25i (=25 interlaced frames per second) on his DVD would preserve the motion smoothness, and he should understand that the corresponding 50 fields per second do not speed up the video but double the temporal resolution = improved motion smoothness. When he insists on 25p (=25 progressive frames per second) for whatever justified or unjustified reason he will have to accept a certain 'stroboscopic' effect for scenes with motion - unless he would artifically increase the motion blur by means of a filter to make it visually appear smoother but reducing sharpness at the same time, not exploiting the potential and quality of the capture. It can be done, but it's not really recommended.

Still, this has nothing to do with the dotcrawl and rainbows of his capture examples, as both fields are equally affected. The reason for this may be his composite video, poor luma/chroma filtering, poorly screend cheap cables producing crosstalk, or it may be baked onto his tape.

[lollo]

^
||

Buy that man a beer!!!