Filters for restoring old VHS

[elbenno]

Dear all, I have been using Virtualdub and Avisynth now for 3 years, so I am not veteran but have some experience.

I want to restore digitally imported old VHS tapes (15 to 20 years old). I like to deinterlace the footage and do some color/artifact correction before encoding them to H264 MKV. As I have quite a few videos to go (around 40 tapes of 2 to 3 hrs) I tested out several settings in both programs, to make a standard filter workflow that I sort of can apply to all tapes.

I first did some runs in Virtualdub, but since I make an intermediate huffyuv encoded vid which I then encode to H264 with a better encoder, my preference now goes to using Avisynth, to do all at once without big intermediary files.

I am importing the videos with a Sharp VC-M31 VHS player. I originally imported the videos in .mpg of around 7 Gb for 3hrs video. Is it worthwhile to import the videos in huffyuv .avi, guess the quality is better then?

After fiddling with lots of filters and deinterlacers in virtualdub, I noticed that (surprisingly) the simple standard blend deinterlace gave me the best result: when I used HQ deinterlacers as smart/smooth/area based deinterlace, I noticed noise/artifact gain, especially some nasty horizontal lines:



I have included 4 sample videos in this link. (Samples with the noise to a black background, sample of the horizontal lines, sample later in the video). Below 2 example screens...



I am now testing Avisynth deinterlacers and mostly get the same (enhanced noise) result. Of course I prefer fast scripts, but quality is more important to me, so slow scripts are welcome. I am also trying to get TempGaussMC_beta2 or QTGMC to work, but these deinterlacers will not work, when I set up my script and test it in AvsP, it gives me an 'function not found' error. I am googling a lot to find answers, but no solutions yet…

Could somebody kindly advise me of avisynth filters/deinterlacers that could help me with correcting these videos, especially the noise and horizontal lines? Many thanks!

[smartel]

capture and keep them interlaced , that's what i would do

capture them huffy and https://forum.videohelp.com/threads/170167-AviSynth-2-5x-and-Convolution3D-guide-for-AVI-captures

[lordsmurf]

Stop deinterlacing.

Read this: Understanding Your Video Source

[elbenno]

Thanks for the suggestions, I will try Convolution3D tomorrow!

Both links were interesting reads, I used to deinterlace all my material, but now I am in doubt... will first check Convolution3D on my material.

[themaster1]

- Capture with the maximum brightness to begin with (between + 16-20 in my capture card compared to default setting). Capture losslessly (lagarith/huffyuv)

- Start with lum,cont,hue tweaks (for hue between +5 ~ + 7.8 for ntsc vhs depending; for commercial tapes) i have verified with avisynth & sony vegas vectorscopes. Filter: Tweak( avisynth)

- adjust the white/black levels (levels filter, avisynth)

- Filters worth of interest: gradations curves, hue, sat intensity, color mill (vdub), NeatVideo ;
avisynth: QTGMC (if you deinterlace), limitedsharpen.. among others...

[elbenno]

Thanks, TheMaster!

I tried out many of the filters found on the Avisynth Wiki. Tweak and Levels are already in my script, nice corrections. Tried out Neat Video, which was ok, maybe a bit to plastic. Will encode with Convolution3D filter and post some results in the coming days...

I will capture with huffyuv/increased brightnes the next time. It's done via USB port, and I now capture via Vdub. Any other suggestions concerning the footage and how to improve it, is greatly appreciated!

[sanlyn]

Tried out Neat Video, which was ok, maybe a bit to plastic.

Never use NeatVideo's default settings. Never. Use it in Advanced Mode and learn to tweak the filter controls.

[poisondeathray]

recapture it, you have macroblocking - this indicates you would benefit (mpeg2 compression wasn't sufficient) . You could also fix the levels when capturing (black level seems too high is the biggest issue)

removedirtmc to get rid of transient defects like comets/dropouts, lines

chubbyrain2 for the alternating pink lines/splotches

mflowinter (from mvtools2) to replace completely bad frames or fields e.g. 2nd frame in clip2

chroma noise filters e.g. fft3dfilter (with chroma plane settings), cnr2

apply a cropped overlay with smoothuv to fix the fixed pattern noise (FPN) persistent left color "rainbow" strip

underlying luminance FPN as demonstrated by your black screen is difficult to treat, but some of the above filters with bringing down the black level will reduce visibility, you might not need other filters like NeuralNet

[jagabo]

Keeping your videos interlaced is the best advice. But...

I am also trying to get TempGaussMC_beta2 or QTGMC to work, but these deinterlacers will not work, when I set up my script and test it in AvsP, it gives me an 'function not found' error.

Did you download and install all the filters that QTGMC() relies on?

http://forum.doom9.org/showthread.php?t=156028

Core plugins
- MVTools2 (2.5.11.2 or above)
- MaskTools v2 (recommend 2.0a45 or above)
- NNEDI3 (recommend 0.9.2 or above)
- RemoveGrain + Repair (extract only the SSE2 versions, some SSE3 versions have a bug and give no real benefit)
- SSE2Tools for YUY2 support (take only SSE2Tools.dll from this zip - do not take the SSE3 version nor the RemoveGrain or Repair dlls)

Additional plugins depending on settings
- [NNEDI2, NNEDI, EEDI3, EEDI2, TDeInt] - if selected directly or via a source-match preset
- Yadif - for Preset="Ultra Fast" or if selected directly (this plugin cannot be autoloaded, it must be loaded in the calling script)
- VerticalCleaner - for SVThin or lossless modes
- AddGrainC (or here) - for Preset="Placebo" or NoiseDeint="Generate" selected for noise processing
- FFT3DFilter - if selected for noise processing
- dfttest - if selected for noise processing
** For FFT3DFilter and ddftest you also need the FFTW3 library (main page / windows zips). On Windows the file needed for both is libfftw3f-3.dll. However for FFT3DFilter, it needs to be called FFTW3.dll, so you will need two copies and rename one. On Windows put the files in your System32 or SysWow64 folder (which depends on your OS and the version of AviSynth that you're running)

VirtualDub has problems with interlaced YV12 when importing MPG video (unless your force the color depth to YUY2). There's also something wrong with the chroma channels in your MPG samples. So they have been mistreated somewhere before becoming MPG files.

Also consider getting a Y adapter cable and splitting the mono audio to both the left and right channels of the capture device.

[elbenno]

Thank you for your tips on filters, poisondeathray, I am sure I will find some nice use for them for this video. Will probably recapture the video, huffyuv avi import should indeed already increase quality...

Jagabo; I was expecting I missed one of these files, already checked once, but will check again.

I will test the filters coming evenings and report back to you all on the progress! Many thanks for your suggestions again, guys

[poisondeathray]

There's also something wrong with the chroma channels in your MPG samples. So they have been mistreated somewhere before becoming MPG files.

Yes those chroma channels are messed up. I wonder what causes it ?

Those pink lines coalese and you get a flickering when you bob it: 2pink, 2non , 2pink, 2non.... .

But the other clip3 doesn't exhibit that problem, so it's not a persistent hardware issue




@ elbenno

Anyways, the filters mentioned above can treat it fairly well....

I attached a comparison file below , left is bobbed, right is filtered

mflowinter often does a decent job of replacing completely damaged frames. If you examine frame by frame the attached file, you will notice damaged frames 4 & 5 have been replaced by fairly good unique frames (non dupes) .

removedirtmc does a good job of any transient defects (including comets , dropouts etc..) - if you look at the attached screenshot, it's even removed (and filled in) the horizontal defect

This example was bob deinterlaced, but you can apply filters and keep the interlacing , it just requires slightly different technique (apply to even/odd grouped fields, then re-weaving)

But a lot of this might be moot - since you should be able to get improved results with a proper capture

[elbenno]

Wow, poisondeathray, very nice! Those filters even got rid of that noisy horizontal line, quite impressive! I will test out those filters as well tomorrow when I am behind my computer. I do not have the originals at my place at the moment, so have to take them at my parents place in a few weeks. It already gives me some hope, watching the far less flickering video that you produced. When I am importing my next videos lossless it might even be a bit better. Thanks!

[jagabo]

Here's an example of the chroma problem:



On the left is one frame. On the right is the next frame. You can see that the hand has moved but the colors for the hand have stayed in the same place.

[poisondeathray]

jagabo - I wonder if that is related to the flashing pink lines?

if you assumetff().separatefields() , each consec. field pair that has the pink, is followed by a gap of orange in roughly the same shape as the defect, as if the chroma was lagging behind a field (or frame if you bob it) . 2pink, 2oranage, 2pink, 2orange...

[elbenno]

poisondeathray, would you mind sharing your script or settings for the filtered example?

[poisondeathray]

No problem, but I suggest you read up on the filters and do some testing to learn how they work - because you will have to change the settings if you recapture, and different scenes will require different settings . For example, your clip #3 has different characteristics (doesn't have the pink flashing lines), so you wouldn't use the same filters or settings

If you understand what the filters are doing, then you can apply the same techniques to treat other parts. For example , there seems to be a consistent green patch in the right upper corner - you should be able to apply the overlay with some filters to treat that. A good tool to use is Avspmod - you can compare the effects of filters and different scripts (push f5 to preview, and you can hit the number keys to look at scripts in different tabs)

I included both 50p and 50i scripts. There are some special rules for handling interlaced content, you have restrictions on cropping for example. QTGMC does some denoising, so you can reduce filters settings in the progressive script (for example, deblocking was elminated cpu=0 instead of cpu=3)

As usual adjust to your tastes

[lordsmurf]

Some of this could be avoided by simply using a better VCR.
Read this: VCR Buying Guide (S-VHS, D-VHS, Professional) for restoring video

What you're using is not very good.

[jagabo]

Some of this could be avoided by simply using a better VCR.
Read this: VCR Buying Guide (S-VHS, D-VHS, Professional) for restoring video

What you're using is not very good.

Yes, that's where the quality starts.

[elbenno]

I am going to first recapture the video over 2 weeks, I will report back to you all with some new screens to discuss a fitting filter script! Thanks already for the nice tips!

[LSchafroth]

Stop deinterlacing.

Read this: Understanding Your Video Source

How does that relate to todays televisions? Here is a sample I found when reading about new televisions:

Deinterlacing

• Plasma and LCD TVs cannot display an interlaced signal. The only devices that actually can are old CRT televisions. To show an interlaced signal, a modern HD flat panel TV has to deinterlace the signal first, combining each pair of frames to produce a progressive signal. Most TV and media comes in the outdated interlaced format, unfortunately, and as a result many flat panel TVs can only accept signals in that format.
  
  

I have converted a VHS video from an old camcorder from the 80s. I converted it to DVD as interlaced. Looked terrible on my 46" Plasma TV. I converted it by deinterlacing it with Virtualdub yadif and burned to a DVD. Looked perfect on the same TV.

This confused me as I've always heard the same thing. Never deinterlace. So why does the deinterlaced look WAY better?

Lannie

[sanlyn]

Something amiss with your plasma TV. Any TV can display interlacd material (but of course newer TVs deinterlace in the process). Retail NTSC DVD and VHS is usually interlaced or progressive+3:2 pulldown. With home-made VHS recordings, the original broadcast can vary. Take another look at your source VHS. Is it interlaced, or is it telecined? Deinterlacing telecined video is a no-no. http://neuron2.net/faq.html

[manono]

Something amiss with your plasma TV.

Or with his process. If, for example, he cropped and/or resized before reencoding it, that could easily have screwed it up.

Before accepting any claim that keeping it interlaced was an inferior solution, I'd like to see samples of the before and the after.

[jagabo]

Yes, all progressive TVs deinterlace or IVTC before displaying the picture. Quality varies from TV to TV. Just as quality of a software deinterlace varies depending on the algorithm used.

The issue is this: if you deinterlace in software now you will forever be stuck with the quality of that deinterlacing. If you leave your video interlaced you will have preserved the quality of the original video, and as the quality of TV deinterlacers improve so will the quality of the interlaced video viewed on it.

[poisondeathray]

If he meant dvd-video, playable in a regular dvd player , I don't see how yadif in vdub could have produced better results on true interlaced material

If he double rate deinterlaced 59.94p, it wouldn't be compliant for dvd-video; and if it was single rate deinterlaced to 29.97p he would be missing 1/2 the information .

Even the worst, cheapest TV deinterlacer (simple bob, no additional processing) isn't much worse than yadif and in some respects it's better

Something doesn't add up - and I'm with manono - I suspect he made some mistakes on making the DVD

[LSchafroth]

Something doesn't add up - and I'm with manono - I suspect he made some mistakes on making the DVD

Thats what I am wondering. If something is amiss in my setup. The tests I did was no cropping, no filtering of any kind. Just deinterlace one and not the other. The latter looks the best every time. This is on a CRT 32" TV and a 46" LG 1080i Plasma TV.

I've been using Pro Coder 3. Should I be using something different?

I've been capturing with a Canopus ADVC-100 via a firewire to my PC.

Lannie

[NJRoadfan]

Here's an example of the chroma problem:


[Attachment 6746 - Click to enlarge]

On the left is one frame. On the right is the next frame. You can see that the hand has moved but the colors for the hand have stayed in the same place.

I've run into this problem with some 8mm tapes. I thought it was artifacts of the player's temporal noise reduction, but the chroma ghosting remained even with it off. It seemed to be embedded in the tape as the problem occurred with multiple playback devices.

[jagabo]

Here's an example of the chroma problem:


[Attachment 6746 - Click to enlarge]

On the left is one frame. On the right is the next frame. You can see that the hand has moved but the colors for the hand have stayed in the same place.

I've run into this problem with some 8mm tapes. I thought it was artifacts of the player's temporal noise reduction, but the chroma ghosting remained even with it off. It seemed to be embedded in the tape as the problem occurred with multiple playback devices.

The most common cause of this problem is handling interlaced YV12 as if it was progressive. For example, if you open an MPG file in VirtualDub and apply any filter that requires RGB, VirtualDub will treat the interlaced chroma channels as progressive when it converts to RGB. That will generate this chroma problem.

[NJRoadfan]

The most common cause of this problem is handling interlaced YV12 as if it was progressive. For example, if you open an MPG file in VirtualDub and apply any filter that requires RGB, VirtualDub will treat the interlaced chroma channels as progressive when it converts to RGB. That will generate this chroma problem.

I'm seeing it with raw YUY2 encoded HuffYUV captures with no filters and outside of VirtualDub. I'll run a few more tests when I get time.

[Kosketus]

I daresay this is a totally dumb comment, but I'll risk the flaming!

If the only TVs sold today are progressive devices, doesn't that mean that anyone who uses a TV to watch - for example - captured VHS tapes (interlaced, by definition) is going to be seeing them deinterlaced, for good or ill? That being so, why does it seem to have become a taboo in some quarters that interlaced source-material must never be deinterlaced? It's going to get deinterlaced anyway before you can watch it, and there's not a damn' thing you can do about that.

The injunction used is "understand your source", as if that alone could determine everything that followed and everything was equally open to choice. But it isn't:- where interlaced sources (specifically video captured from broadcasts, either analogue to videotape or DVB to hard disk) are concerned the only question is whether to deinterlace it oneself or leave it to the TV to do it. At the end of the chain whatever is fed into the TV will (unless it's already been - or was from the start) get deinterlaced.

Seems to me that anyone who has the skills or wants to develop them is going to prefer to do the deinterlacing themselves, whereas someone like me (ie who's as clueless as me about doing deinterlacing well) would do better not to try but instead to let the TV do its stuff - on the assumption that the result is unlikely to be worse. But that's a different argument.

[lordsmurf]

I daresay this is a totally dumb comment, but I'll risk the flaming!

If the only TVs sold today are progressive devices, doesn't that mean that anyone who uses a TV to watch - for example - captured VHS tapes (interlaced, by definition) is going to be seeing them deinterlaced, for good or ill? That being so, why does it seem to have become a taboo in some quarters that interlaced source-material must never be deinterlaced? It's going to get deinterlaced anyway before you can watch it, and there's not a damn' thing you can do about that.....erlaced.

Seems to me that anyone who has the skills or wants to develop them is going to prefer to do the deinterlacing themselves, whereas someone like me (ie who's as clueless as me about doing deinterlacing well) would do better not to try but instead to let the TV do its stuff - on the assumption that the result is unlikely to be worse. But that's a different argument.

You're missing the point.

Deinterlacing improves with every generation. If you hard deinterlace the source, it's going to look that way forever. And unless you're running something like QTGMC, or even a proprietary method from Snell & Willcox or Faroudja, then what you're doing is guaranteed to look like crap.

The ideal solution is to leave it alone, and let the hardware handle. For starters, most hardware is pretty good at deinterlacing. Interlaced footage will look better with each generation of hardware. It does improve every 3-5 years.

It has nothing to do with "skills" and everything to do with the available technology. Those who claim they have better deinterlacing "skills" are quite frankly talking out of their ass. At best, somebody is simply claiming to know how to use Avisynth, for a fairly well-documented filter (unlike other filters, which is my main beef with Avisynth).

Therefore it has nothing to do with "taboo", and everything to do with retaining optimal quality. Deinterlacing for TV is fixing something that wasn't broken to begin with. (And said "fix" is actually breaking it.)