10-11-2024, 02:57 PM
|
|
Free Member
|
|
Join Date: Oct 2018
Location: Belgium
Posts: 32
Thanked 2 Times in 2 Posts
|
|
Hi all,
After having captured more than 100 VHS cassettes and learning and improving with every capture, this is the first time I'm actually considering restoring one - the reason being the quality of the program and (possibly) its rarity. However there are a lot of factors that make it (at least for me) difficult to get started, and even question if it's worth doing. I hope I can get some advice on how to continue!
THE SOURCE
To begin with, I already knew this video for over a decade because it was on YouTube ( https://youtu.be/a-kznTN66Ho), but I almost couldn't watch it because of the bad quality. Last month, I found a cleaner copy on a DVD-R in a jazz collection, and also the VHS from where it was sourced. The DVD-video proved to be a considerable upgrade, but I was suspecting that I could get it even better by doing a new VHS capture myself. But that's where I already got stuck: I cannot seem to get the fresh capture much better than the DVD video. Here is a still from the DVD, compared to captures I did myself using combinations of different machines:
DVD
compare - jarret.01DVD.jpg
left: Panasonic NV-FS200 + DataVideo TBC3000 // middle: JVC HR-S7600 + DataVideo TBC3000 // right: JVC HR-S7600 + Panasonic DMR-ES10
You must be logged in to view this content; either login or register for the forum. The attached screen shots, before/after images, photos and graphics are created/posted for the benefit of site members. And you are invited to join our digital media community. |
DVD
compare - jarrett.01.jpg
left: Panasonic NV-FS200 + DataVideo TBC3000 // middle: JVC HR-S7600 + DataVideo TBC3000 // right: JVC HR-S7600 + Panasonic DMR-ES10
You must be logged in to view this content; either login or register for the forum. The attached screen shots, before/after images, photos and graphics are created/posted for the benefit of site members. And you are invited to join our digital media community. |
I’ve attached the stills, and ~5 second snippets from all 3 captures for real comparison. I captured in FFV1 using VirtualDub2 on Windows 7, but converted the snippets to Lagarith.
What capture has the potential of improving on the DVD version? I'm not exactly sure on how the DVD version was made, but I believe on a JVC S-VHS player, without external frame TBC. I have no idea of the capture card, and when the capture was made (could be 10-15 years ago). Is it possible that the VHS has degraded so much over time that a better capture is not possible? I can see visually that the tape is not very well rolled up around the spool, which could explain some of the differences between each capture. Or is there something wrong with my capturing process?
INTERLACING
What I figured out was that the video is actually progressive, with each bottom field corresponding with the top field of the next frame. To fix this I shuffled a bit in AviSynth+:
Code:
vid=LWLibavVideoSource("D:\MDJ-keithjarrett[PANA-TBC3000-ATI600].avi")
F1=SeparateFields(vid).SelectOdd()
F2=SeparateFields(vid).SelectEven().Loop(2, 0, 0)
vidfix=Interleave(F1,F2).Weave()
RESTORING
Although I've read a lot about restoring and filtering (mostly on this forum), I'm really a bit oblivious on how to start something like this. The video appears in bad shape and has a lot of problems - some of them coming from the fact the program was shot on film, some from the fact that it’s a multi-gen VHS copy. The first step would be to detect and name all the defects and artifacts that are present in the video, then to find appropriate filters to correct them. I was hoping I could get started by asking some advice here.
My own thoughts:
I seem to like the picture of the JVC in combination with the TBC3000 the most, and also how it handles the multiple dropouts. Correct me if I’m using the wrong terms: the ES10 introduces more chroma noise than the TBC3000; also the color banding at the top is least obvious with the JVC in combination with the TBC3000. The Panasonic seems to have a hard time getting the color from the VHS, it is quite dark with a lot of chroma noise. Soundwise I’ve always preferred the Panasonic over the JVC though.
Another complication I just found out about : when my monitor Sony TV is turned on, it generates a flickering wave (an upwards rolling bar of luminance). I believe all my captures were made with the TV turned on, so this flicker is probably baked in all the captures. I tried doing another capture today, but I have the feeling the quality of the tape is already degrading (more inserted frames, worse chrome noise…)
Thanks a lot for taking the time!
Karel
P.s. If anyone knows more about this broadcast, let me know. I would assume it to be an American TV broadcast, but the presenter’s accent sounds a little bit Australian? If I could pinpoint the network/station, I could inquire if a master tape still exists, and then all this work is possibly unnecessary.
|
Someday, 12:01 PM
|
|
Ads / Sponsors
|
|
Join Date: ∞
Posts: 42
Thanks: ∞
Thanked 42 Times in 42 Posts
|
|
|
10-11-2024, 04:31 PM
|
|
Free Member
|
|
Join Date: Sep 2020
Location: Perth, Western Australia
Posts: 640
Thanked 122 Times in 110 Posts
|
|
The YT copy looks 2nd gen or worse, perhaps LP or EP. Terrible linear audio. Dirty probably 16mm film source which film having an optical sound track is probably at least 2nd gen before it was transferred to videoed.
There's really no substitute for clean first gen sources, in this case the original 16 mm camera films plus the magnetic sound elements including a modern HD scan of the film and transfer from the probably Nagra location audio tapes. That is what a pro would do (if those elements were still available and in good enough shape). What the public may believe is a magical "restoration" may just involve having access to well preserved original picture and sound elements and expertly using them to their best advantage using good, well maintained gear.
Last edited by timtape; 10-11-2024 at 04:59 PM.
|
10-12-2024, 12:55 AM
|
|
Free Member
|
|
Join Date: Jul 2023
Posts: 552
Thanked 77 Times in 75 Posts
|
|
The YouTube video does seem pretty popular with 116K views, so seems worthwhile to put a better copy out there. Do you know what the DVD recorder chain was when it was first made?
There definitely are some avisynth wizards here that could probably work some additional magic on whatever your capture comes out to be.
|
10-12-2024, 01:50 AM
|
|
Free Member
|
|
Join Date: Feb 2022
Posts: 105
Thanked 24 Times in 22 Posts
|
|
Attached a sample what (using Vapoursynth) Spotless + BasicVSR++ + FilmGrain does to the first sample just do give a general feel,...
|
The following users thank Selur for this useful post:
lordsmurf (10-12-2024)
|
10-12-2024, 11:04 AM
|
|
Free Member
|
|
Join Date: Oct 2018
Location: Belgium
Posts: 32
Thanked 2 Times in 2 Posts
|
|
My tape is also at least 2nd gen: it's a compilation of jazz broadcasts from Swedish television, probably taped of the air in Sweden and then somehow found their way on a single VHS copy to Belgium.
Finding the original 16mm film, or TV station master videotape would of course be much better solution, but this is one of the rarest pieces I've come accross. I've got quite a good overview of what exists (or once existed) of jazz TV shows, but I can't find anything about this broadcast - also because I can't pinpoint the American/Australian channel which made the program in the first place.
Anyhow knowing how much from the seventies was destroyed or taped over, I don't have much hope of founding the source material...
Quote:
Originally Posted by timtape
The YT copy looks 2nd gen or worse, perhaps LP or EP. Terrible linear audio. Dirty probably 16mm film source which film having an optical sound track is probably at least 2nd gen before it was transferred to videoed.
There's really no substitute for clean first gen sources, in this case the original 16 mm camera films plus the magnetic sound elements including a modern HD scan of the film and transfer from the probably Nagra location audio tapes. That is what a pro would do (if those elements were still available and in good enough shape). What the public may believe is a magical "restoration" may just involve having access to well preserved original picture and sound elements and expertly using them to their best advantage using good, well maintained gear.
|
|
10-12-2024, 07:09 PM
|
|
Free Member
|
|
Join Date: Sep 2020
Location: Perth, Western Australia
Posts: 640
Thanked 122 Times in 110 Posts
|
|
Quote:
Originally Posted by OBNOXIUs
My tape is also at least 2nd gen: it's a compilation of jazz broadcasts from Swedish television, probably taped of the air in Sweden and then somehow found their way on a single VHS copy to Belgium.
Finding the original 16mm film, or TV station master videotape would of course be much better solution, but this is one of the rarest pieces I've come accross. I've got quite a good overview of what exists (or once existed) of jazz TV shows, but I can't find anything about this broadcast - also because I can't pinpoint the American/Australian channel which made the program in the first place.
Anyhow knowing how much from the seventies was destroyed or taped over, I don't have much hope of founding the source material...
|
Fair enough, although if the program was shot on 16mm film that's more likely to have survived as unlike videotape it couldnt be erased for reuse. Many live and originally videotaped programs survive on 16mm only.
I just checked out your video upload samples. Obviously the basic picture is a significant improvement on the YT upload,
but I couldnt hear any audio on any of the three samples.
Possibly your VHS tape also has, potentially at least, the best audio available. Any chance of uploading a sample of that? If it has both linear and HiFi tracks, hearing each separately would be good. Thanks.
Last edited by timtape; 10-12-2024 at 07:49 PM.
|
The following users thank timtape for this useful post:
lordsmurf (10-12-2024)
|
10-12-2024, 07:10 PM
|
|
Premium Member
|
|
Join Date: Jan 2017
Location: USA
Posts: 202
Thanked 92 Times in 66 Posts
|
|
A vectorscope such as VirtualDub Color Tools plugin or AviSynth Histogram in color or color 2 mode shows JVC has many random bursts of color noise coming and going from frame to frame. Although most of that can probably be removed with denoisers such as Spotless or RemoveDirtMC or TemporalDegrain, etc. My guess is the TBC/DNR function in the Panasonic has removed most of that color noise, were as JVC hasn't.
For JVC TBC3000, AviSynth Histogram("levels") shows the video barely going outside the 235 upper limit on the luminance. Nothing major. But a reminder when using TBC-3000 make sure to use the Proc Amp to keep the Luminance in range (in Virtualdub Preview Histogram), since the ATI 600 doesn't have internal proc amp controls.
Panasonic has more muted color and contrast. The TBC-3000 Color, Brightness, and Contrast controls can change all of that. The JVC and ES10 combo also has more muted contrast, probably because the ES10 is known to darken things. Avisynth Levels function can adjust luminance to the full gamut and also shift the gamma to bring more of the shadow into the midrange. I do my best to capture things with proc amp settings that use most of the 16 thru 235 legal gamut for Luminance, but try to leave some headroom on high end because it can be adjusted later with Levels function and I'd rather not crush details during capture that I cannot undo later.
With multigen especially, color correction is usually needed. In this case, the flesh tones have a green cast. If you are going to get into contrast and color adjustments with Avisynth or whatever, then JVC doesn't necessarily have an advantage. Panasonic may be easier to work with because the color isn't as saturated and so it could be corrected, then you can increase the saturation as much as you'd like. On the other hand, if JVC has more saturated color but is inaccurate (multigen copy), you may have to turn some colors down more if they are inaccurate and not correctable. Panasonic tends to preserve more detail in the luminance (Y channel) than JVC which tends to smooth more of that out. It's the eternal debate between whether you prefer softer picture or more details (as well as the more muted color of Panasonic vs more saturation of JVC). However, when dealing with multi gen tapes such as this, much of that detail is already lost. So JVC may not make much difference on softening over the Panasonic.
I did a quick and dirty take on the Panasonic clip. Attached is avs script I used (Avisynth 2.6 32-bit, if it matters) and resulting clip. I also added Camcorder Color Denoise (CCD) plugin in Virtualdub with default settings. Don't have time to break it all down. Tried to comment things in the file. This is all stuff that's been discussed many times on the forum. I learned it all from Sanlyn's posts.
I didn't get into anything like de-telecine (haven't played with that stuff) or de-interlacing. If this content is mixed film telecined to video and native video, then I suppose de-telecine doesn't really work, and deinterlacing is the only other option. Although, you can use QTGMC and then re-interlace the file, or separate the fields as was done in this script for the Spotless work and use the same denoisers QTGMC uses.
In general, all of these different caps seem rather similar in how much detail is preserved. Most of the remaining issues involve denoising and enhancing contrast/color. Those are things that have to be done in post. The most important thing is to do no harm during capture. If you can fix it in hardware, do so. Not everything can be fixed in post.
P.S. - I'm a jazz fan, and this is a wonderful piece of video and audio. Thanks for what you are doing to try and track down the best source you can and create the best copy you can. The good news is the existing clip on YT puts the bar on the floor. So you'll be successful no matter what! I spend most of my free time with my own projects in the same spirit of trying to preserve/restore as best I can content that is otherwise lost or I can only find in poor quality. Best of luck to you!
|
The following users thank keaton for this useful post:
lordsmurf (10-12-2024)
|
10-12-2024, 08:12 PM
|
|
Site Staff | Video
|
|
Join Date: Dec 2002
Posts: 14,182
Thanked 2,576 Times in 2,188 Posts
|
|
Quote:
Originally Posted by Selur
Attached a sample what (using Vapoursynth) Spotless + BasicVSR++ + FilmGrain does to the first sample just do give a general feel,...
|
I can't open this. Neither VirtualDub2 nor VLC.
Code:
Codec not supported:
VLC could not decode the format "av01" (No description for this codec)
It's not a widely-adopted standard yet, and may never be.
EDIT: Chrome opens it, no media player needed. Hmm, it seems a bit plastic in this sample. Pull back on the NR.
|
10-13-2024, 03:26 AM
|
|
Free Member
|
|
Join Date: Feb 2022
Posts: 105
Thanked 24 Times in 22 Posts
|
|
Plays fine in MPC-HC and other libav based players. (like latest vlc-4.0.0-dev-win64-6210ebe9)
=> maybe update your vlc,...
Quote:
Hmm, it seems a bit plastic in this sample. Pull back on the NR.
|
That's what BasicVSR++ does to the source, not much to do about it.
Since the source didn't really show many details to begin with, I did no special tweaking and just used plain BasicVSR++ (no masking and other filters).
|
10-13-2024, 09:44 AM
|
|
Free Member
|
|
Join Date: Oct 2018
Location: Belgium
Posts: 32
Thanked 2 Times in 2 Posts
|
|
Thanks Selur. The AI upscaling indeed makes the whole thing a bit glassy. I guess it could work possibly as a final stage, when first the VHS artifacts and then the film stains are removed? I will check out Spotless first..
I know that denoising and color correction are the obvious ways to improve the footage, it's just that I have no experience at all doing stuff like that. So it's nice to know what filters more experienced people use to get me started.
|
10-14-2024, 06:10 AM
|
|
Free Member
|
|
Join Date: Oct 2018
Location: Belgium
Posts: 32
Thanked 2 Times in 2 Posts
|
|
Quote:
Originally Posted by keaton
I did a quick and dirty take on the Panasonic clip. Attached is avs script I used (Avisynth 2.6 32-bit, if it matters) and resulting clip. I also added Camcorder Color Denoise (CCD) plugin in Virtualdub with default settings. Don't have time to break it all down. Tried to comment things in the file. This is all stuff that's been discussed many times on the forum. I learned it all from Sanlyn's posts."
|
Thanks keaton for the script, that’s a big improvement on my original capture!
Interesting that part of the livelier (more saturated?) color from the JVC is in fact just false information stemming from it being a multigen copy. It tricked my eye at least.
I had some technical issues getting your script working at home, but finally found a solution. But for the sake of documenting and helping others (and myself) out in the future I’ll list my problems here. I’m using VirtualDub2 (x86) with AviSynth 2.6 32-bit ST: - I couldn’t even open my own uncompressed sample with AviSynth. The picture was gray in VirtualDub with the following error: “Avisynth read error: CAVIStreamSynth: System exception - Access Violation at …”. But compressed with Lagarith or HuffYUV there were no errors. Same happens in VirtualDub 1.9.11.
CAVIStreamSynth.jpg
- But then I removed the Spotless function, and the problem went away. Which made me think I had a colorspace problem. And indeed when using AVISource() in combination with ConvertToYUY2(interlaced=true) the problem went away. Turns out when I exported the snippet to use as a sample here on digitalfaq, I used uncompressed AVI in this format:
Code:
Format : YUV
Codec ID : YV16
Codec ID/Info : Elecard YUV 4:2:2 Planar
Color space : YUV
Chroma subsampling : 4:2:2
Compressing to HuffYUV or Lagarith fixed this issue.
But when I tried using my FFV1 compressed full capture, I got the same error.
- While Spotless wasn’t working on my FFV1 source, I looked for other denoisers and found SpotRemover and SpotLess_Mod, which claim to preserve more detail in fast moving images than SpotLess. I thought that would be a good option for fast moving fingers on a piano. SpotRemover throws the following AVISynth error: “DeSpot: Input to filter must be YV12 or YUY2 planar!”. SpotLess_Mod gave an error in VirtualDub: “Avisynth read error: Invalid memory alignment. Used unaligned crop?”. When I performed the noise reduction on the full frame instead of the fields independently, the error went away. So my script now looks like this:
Code:
function clean(clip c, int size) {
a=last
AssumeTFF(c)
SeparateFields()
e=SelectEven()
o=SelectOdd()
Interleave(e,o.loop(2,16,16))
Weave().SpotLess_Mod(BlkSz=size)
}
FFmpegSource2("MDJ-keithjarrett[PANA-TBC3000-ATI600].avi", atrack=-1).trim(45,0).DelayAudio(-0.370)
So is there a reason that SpotLess is used on fields separately, or can I just as easily perform it on the full frame (after the fields are aligned)? But I guess easier is just using ConvertToYUY2(interlaced=true) on the FFV1 source, since this introduces no real color loss?
Another question about the restoration: there is no need to align the chroma (chroma offset)?
Quote:
Originally Posted by keaton
For JVC TBC3000, AviSynth Histogram("levels") shows the video barely going outside the 235 upper limit on the luminance. Nothing major. But a reminder when using TBC-3000 make sure to use the Proc Amp to keep the Luminance in range (in Virtualdub Preview Histogram), since the ATI 600 doesn't have internal proc amp controls.
|
I tried keeping my captures in official luma levels, but indeed I sometimes let it go a bit above the upper limit. My reasoning being that the information is not lost - I can always correct it afterwards - but also thinking that those high luma values are possibly coming from noise and not actual video content? Also I don’t want to go through the whole tape and wear it down further just to set my levels…
So I set my ATI600 levels on default, except: brightness=121, sharpness=0, and then try to fill the whole luminance range by tweaking the TBC3000 proc-amp.
The capture to DVD from some years ago had some ‘soft’ quality that I liked. It seems to come from the scan lines not being perfectly aligned with each other, resulting in rough edges. Could it be that the TBC independently working on the bottom field of the current frame and the top field of the next results in shaky lines when combined? I can maybe try horizontal shifting one of the fields and see if it improves. Now I randomly tried to add the VDub filter DFN VHS (after CCD) to make the edges softer and I kind of like the result. I played the restored Panasonic capture side-by-side with the DVD, and the result is much more pleasant to watch. What’s striking is how many more drop-outs/warbled tape moments my capture has vs the DVD… It’s unfortunate I didn’t get my hands on the tape 10-15 years ago.
Thanks!
Karel
|
10-14-2024, 11:01 PM
|
|
Premium Member
|
|
Join Date: Jan 2017
Location: USA
Posts: 202
Thanked 92 Times in 66 Posts
|
|
Quote:
Originally Posted by OBNOXIUs
Thanks keaton for the script, that’s a big improvement on my original capture!
|
Great! The idea was to give you something quick to show a rough idea of what Avisynth can do with just some color/contrast adjustment and a denoiser. With video like this, it's going to take different tools for different parts of the video, and different settings for different shots, but it's a quick sampler to give an idea of whether or not to proceed.
Quote:
Originally Posted by OBNOXIUs
[*]I looked for other denoisers and found SpotRemover and SpotLess_Mod, which claim to preserve more detail in fast moving images than SpotLess.
|
Great! Hope it works out for you. Definitely something to be mindful of with this video. There are so many denoisers out there. So take your time and experiment if you can afford to. I wish I knew more about denosiers. They're amazing tools, when they work right, and many have so many parameters to tweak if you wish.
Quote:
Originally Posted by OBNOXIUs
So is there a reason that SpotLess is used on fields separately, or can I just as easily perform it on the full frame (after the fields are aligned)?
|
I've recently tried using Spotless when my default RemoveDirtMC isn't strong enough. Much of the time I don't need a heavy denoiser, just something that can get rid of VHS grain and many comets/dropouts. What I've observed after many hours of video with RemoveDirtMC, is I've seen fewer problems with removing non-noise or small moving objects or with just weird glitches on a part of a frame when separating the even/odd fields before cleaning. My impression is that many of these type of temporal denoisers are either designed for or are just better on progressive material. So separating fields achieves this result because there are no "jagged edges" (unless they are "baked in" from bad copying from one tape to the next) like you see with interlaced frames. You can try this for yourself and see the differences when running a given denoiser on a clip and then trying the same clip using the separate fields approach. Sometimes, you may see no difference. Other times, you may see problems with full frame vs not with separated fields. I don't have as many hours with Spotless, so I cannot recall how often this has helped with that.
Your clip didn't seem to get enough denoising from RemoveDirtMC (at least on lower settings like 30 or 40 that are typically as far as you can go before it starts to remove things you don't want it to), so I did this example with Spotless. But how much noise you want is entirely up to you. Too much can make it look like a painting and not life-like. Although there is a function to add grain back ( http://www.avisynth.nl/index.php?title=AddGrainC) that might help it look better if you find it looks too processed because you had to resort to something really strong such as TemporalDegrain.
Also, deinterlacing with QTGMC includes a denoising stage with dfttest or fft3dfilter. That is done after the video is deinterlaced/progressive. So, just another example of denoising being done without interlaced jagged edges. You can also do this yourself without QTGMC by separating fields and running the same denoisers on even/odds separately, if you want to preserve the native interlaced format for archival purposes or for MPEG2/DVD encoding.
Quote:
Originally Posted by OBNOXIUs
Another question about the restoration: there is no need to align the chroma (chroma offset)?
|
Not sure if I understand the question.
If you're referring to ChromaShift ( http://avisynth.nl/index.php/ChromaShift), yes. It's likely you'll discover some bleeding color on the edges of objects in the video due to generational loss. Not always noticeable, but some color combinations will really make it easier to spot. Plenty of examples of ChromaShift use in prior forum threads.
If you're referring to ColorYUV with off_u and off_v arguments, that's up to you. I noticed the peak of the U and V channels in the Avisynth Histogram("levels") output were a bit off from the center line. They don't necessarily need to be aligned that way, depending on what's in the image. It's a strong way to make color adjustments, but that kind of color shifting can be from multigen copies.
One advantage you have with this video is there's a lot of neutral color in the keys and background. So making color adjustments (possibly with color/contrast adjustments in Virtualdub with the ColorMill or Gradation Curves filters, if you're familiar with Photoshop style color correction) to get Red/Green/Blue close to each other in the neutral colors (I use csamp.exe to see the RGB values of pixels, found the exe or a link in this forum many years ago) , then working on the flesh tones (some color correction knowledge helps here, but there's also that line on the Vectorscope in Virtualdub Color Tools between Red and Yellow that is a good guideline for roughly where you want the flesh tones to be. If you crop the picture to just look at some skin, that can help you see where it lines up or if you learn about the RGB rules for flesh tones from color correcting in photoshop, i.e. Red minus Green is greater than Green minus Blue). If I took another stab at this, I'd try that approach, and then use hue adjust in Tweak to try fixing flesh tones if RGB adjustments didn't work. Reason being hue adjustments don't effect the neutral colors. So if you have good shadow/midrange/highlight values that are neutral (Red=Green=Blue or close to that) and other colors still seem off, it can be a sign that the hues need a shift. Use the vectorscope to see the effects of that. You can do selective hue shifts on certain hues in Tweak using the startHue and endHue arguments, but typically I do it for all hues.
Quote:
Originally Posted by OBNOXIUs
So I set my ATI600 levels on default, except: brightness=121, sharpness=0, and then try to fill the whole luminance range by tweaking the TBC3000 proc-amp.
|
That seems fine. When using a Proc Amp in the video chain, I set all the Virtualdub levels to 128.
Quote:
Originally Posted by OBNOXIUs
What’s striking is how many more drop-outs/warbled tape moments my capture has vs the DVD
|
When the tape is inferior to the DVD due to heavy damage, you could try converting the MPEG2 from the DVD to the same AVI codec ( Huffyuv/Lagarith) and use frames from that as needed in Avisynth. Of course, the frame rate, audio sample rate, and resolution need to be the same for this to work.
|
All times are GMT -5. The time now is 08:31 PM
|