VHS Family Wedding tape restoration

[2Bdecided]

I've replied in the other thread, but...


My user name was chosen for a reason. I'm lousy at making decisions. If I had to pick, I'd pick the DVD-R version. As for deciding between edit.norm.soft.sharp I can't even decide that on my own captures and just leave it (the slider, on the machines that have it as a slider) in the middle. There is a "Panasonic edge look" to Panasonic VCRs though. I have a feeling what's on the tape is closer to what you see in the "soft" position.



When you've captured it, you're going to hit it with some script, part of which should be written to reduce the aliasing in the original camera. I think the interaction of that script with the capture will be crucial, and it's the output that you care about. See if it works (or can be toned down, so it works even better) with the new captures, and whether it can be made to work better in one sharpness setting that the other.


I think you'd get a better response to solving the aliasing over on doom9, though there was already a good solution in this thread from sanlyn. Can you do better with sharper or softer captures sanlyn?


Cheers,
David.
P.S. As it's your wedding video, and you're "only" capturing DV, there's no harm in keeping several full captures around "just in case".
P.P.S. I only listened to the audio for the first time today. It's possible to remove half of the hiss without ruining it, but decent audio plug-ins cost a small fortune. Your video was well produced for the time, wasn't it? I guess they didn't allow cameras in church? They'd probably just charge you extra now and be glad of the extra income.

[Tafflad]

Yep .. no camera in the church ... I do have a copy of Soundforge if you are familiar with that, there are presets for VHS tapes, but can't get them to work.
I was pleased with the production, while looking very 80's .... compared to the Best Man .. he got married a year earlier .. and his video was like an episode of Top of The tops ... cuts all flying in from different angles, fades going to polarization & kaleidoscope effects.

[DB83]

What about the 'Wedding Night' video

[Tafflad]

over exposed

[sanlyn]

I think you'd get a better response to solving the aliasing over on doom9, though there was already a good solution in this thread from sanlyn. Can you do better with sharper or softer captures sanlyn?

Softer, no. The plugins and code I used to clean aliasing softens anyway, and you really have to be careful using them, and tweak until you get migraines. Those routines were from Doom9 and jagabo. Of course over sharpening is a nightmare no matter what else is going on.

[sanlyn]

HI Sanlyn ... as it was your original observations that have started me trying recaptures ... do you want me to carry out what I proposed in post #118 or any other mix & match combination.
2Bdecided.. obviously same Q to yourself

I think 2BDecided has made some good evaluations. It might ultimately come down to which method works best with specific videos. The more you watch this stuff the more frustrating it gets, but you learn more about spotting problems.

[Tafflad]

OK ... redid capture off original tape ...
Sample 7

Capture path is: VCR <svideo> DVDR <svideo> ADVC <firewire> PC

VCR is set to picture control 'Mid Position'
Edit set to 'ON'

Capture is with no processor control adjustments, other than setting to PAL.
WINDV set for Type2

[2Bdecided]

You are wearing the tape out. You need to capture it now. Don't delete your previous captures either.

Using EDIT on the latest sample...
1. makes the levels a little better (but you could fix that anyway)
2. increases the saturation a little (but you could do that anyway if you wanted)
3. seems to make this particular picture a little more green (which I think is undesirable, but could be fixed),
4. seems to make the picture a little sharper (but you could do that anyway with sharpen(0.5,0.0) or something better)
but to my eyes...
5. there's also a little more genuine real detail. Hard to tell. I don't think it's just more noise or artificial sharpening. There seem to be more details/folds visible in the fabric than in the previous capture. The VCR seems to be processing the images less, which is a good thing. See if anyone else thinks the same.


If so, the EDIT on position is worth using, as long as it doesn't irrecoverably mess with the colours in some other shot. See what sanlyn says though. He's the colour correction guru.


Cheers,
David.


P.S. quick comparison script - take the sharpen out to make it fairer...

#load sources for comparison
a=avisource("Sample 6 via DVDR.avi")
b=avisource("Sample 7 - Mid+Edit On+DVDR.avi")


#find same frames in all, try to match levels and colours
a=a.Trim(0,754).Levels(0,1.0,255,0,250,coring=false)
b=b.Tweak(sat=0.85,coring=false)


#bob all
a=a.bob().Sharpen(0.5,0.0)
b=b.bob()


#subtitle (label) all
a=a.Subtitle("re-capture via DVD-R", align=3).ScriptClip("subtitle(string(current_frame),align=1)")
b=b.Subtitle("re-capture via DVD-R + EDIT", align=3).ScriptClip("subtitle(string(current_frame),align=1)")


#interlave all
interleave(a,b)


#greyscale()
#histogram(mode="levels")

[Tafflad]

I have very little skill on Avisynth .... if I follow your comment I deleted
.Sharpen(0.5,0.0)

Then run and watch line by line is that right ? one frame one capture next frame the other .... left & right sides of VD show same image ?

The +edit does show more detail.

[2Bdecided]

Yes, exactly.

VirtualDUB shows two images because VirtualDUB itself can also be used to filter and change things. But as you're not using any VirtualDUB filtering (just using it as a viewer/player) both sides show the same.


Cheers,
David.

[Tafflad]

Sanlyn .... do you also concur as well that capture without full frame TBC and the method used in ' Sample 7' is the way to go ....

i.e. Capture path is: VCR <svideo> DVDR <svideo> ADVC110 <firewire> PC

VCR is set to picture control 'Mid Position'
Edit set to 'ON'

Capture is with no processor control adjustments, other than setting to PAL.
WINDV set for Type2

If you agree this is 'better than original 'Sample2' then I'll fully recapture using this method.

[sanlyn]

Thanks for the new capture. Sample 7: aliasing actually looks a little more severe than the earlier cap of the same scene. But it's my guess that it will look a little different every time. The blue discoloration isn't quite as bad (it's still troublesome), but if the tape has aged that way there's really not much one can do. Detail and most color and levels overall look somewhat but visibly cleaner. By saying the detail looks "cleaner" I don't mean that there is more, its just less noisy except for the spastic aliasing. Keeping the skin tones natural would be the way I'd go, and just try as best as one can to get clean whites (when the blue booboo doesn't interfere).

Most of the clay-face and over-filtering is going to result from routines used to clean up the bad edges. I'd use the same doom9 routines used earlier. But I'll still be cruising forums to see if I ferret out better methods that don't smooth everything into oblivion. Adding some noise (fine grain effects) and avoiding too much sharpening is one way to combat posterization. Also, really strong color corrections (not needed here, thank heaven) can contribute adversely to the kind of oil-paint and poster effects that anti-aliasing often generates.

The bluish stain is one matter, but the King Kong in the room is that aliasing. OK, I can accept that it's the camera's problem (or whatever), but I'd say more accurately that Satan did it.

[Tafflad]

Sanlyn - Thnx ... I'll go ahead and recapture whole tape.

Now the 'very nice ask' back earlier in this thread ( post #33) you had got as far as ver 6b of your script.
Would you care to (or be persuaded to) have a look at the script with ref to capture7 (Sample 7 - Mid+Edit On+DVDR)
That will bet he capture chain I will use to redo the tape ... so if any tweaks are needed now there is a different capture 'baseline' would very much welcome your skills on this.
Obviously any of the other skilled guys who have contributed to this thread, feel free to pass any suggestions - I'll get on with recapture.

[sanlyn]

Dun and games in DV capture land. The Sample 7 capture is far too bright. No crushed darks at least, but brights are blown away to smithereens. VHS->DV is difficult enough to clean, nut bright captures also pose gamma problems. Be that as it may:

Sample 7, frame 120 (original). There is motion in this frame:

[Attachment 22787 - Click to enlarge]

Sample 7, after script and VirtualDub. Pasty color, a signature of analog->DV captures:

[Attachment 22788 - Click to enlarge]

Sample 7, frame 472 (original), with humongous aliasing and edge ghosting:

[Attachment 22789 - Click to enlarge]

Sample 7, frame 472 after script + VirtualDub:

[Attachment 22790 - Click to enlarge]

Cleaning up VHS takes a long time. Cleaning up VHS->DV takes a looooooooonger time and destructive filters.

[sanlyn]

Attached, the results of a day of fun and games referenced above. The attached mpg is as far as I could get. Could probably tweak more, but for little gain. The project files.zip attached contains a .vcf file discussed in the script, an Avisynth plugin and an .avs function that I used. This would be a nice sharp capture if not for aliasing and horrid edge ghosts. Setting up levels and color for the mpeg was strum und drang: levels and color change in every shot, and during each shot every time the camera moves. A "pro" using autogain and autocolor? Perish the thought. Remedy: spend the next 18 months correcting each camera shot individually. Excelsior! Critics have at it (that's what the forum's for).

[Tafflad]

Gives me something to have a play around with on weekend ... very much appreciate your efforts ....far far better than I would have got on my own.

[sanlyn]

Color on this one was tricky. Script #1 just interlaces and does some repair (it's still green). Script #2 and VDub did the rest. I somehow managed to noise up the video, so I downloaded a new MPEG to post #135 just now with clean audio.

[2Bdecided]

The Sample 7 capture is far too bright. No crushed darks at least, but brights are blown away to smithereens.

They're not though, are they? There's absolutely no clipping. They're above 235 but below 255. Given that you're going to fix levels and colour correct, the fact they're above 235 is irrelevant. Some would even claim the extra range gives you more precision in the correction (I wouldn't - the amount of noise makes it irrelevant).

Few things about the scripts:


1. DirectShowSource isn't the best choice for a DV source. You should use AVIsource and make sure Cedocida is the DV codec used (typically by it being the highest or only DV codec in the video codecs list in Windows). Cedocida should be set to return DVD/MPEG-2 interlaced YV12 chroma, or YUY2 chroma. You'll need the ConvertToYV12(interlaced=true) line with the latter, but it becomes unnecessary (and don't do anything) with the former.


2. It's amazing that you've made white things white again, and got the dresses right, but I think the skin colours have suffered a little. I'm not looking on a calibrated monitor, and I'm not about to burn it to DVD to check, but I think the pinky skin colours of the original version may be more true-to-life than the browny ones of the corrected version. Something in between is probably correct.


3. I wish I could figure out a way of maintaining some noise on this. The original aliasing is so bad that I don't know how to, but the output is too clean and plasticky for my liking. Don't get me wrong, the removal of the aliasing is near-miraculous, and you've probably kept almost all of the detail that there is (I think I counted 7 filters), but... People can be quite forgiving of VHS, because they think "I'm watching VHS - it looks like this" and then concentrate on the content. I don't think people are as forgiving of images with no faults but no detail. IMO. YMMV. etc. I see the addgrain lines but can't really see it in the output - I don't like artificial grain anyway, so no bad thing.


4. If I've deinterlaced something while restoring, I keep the deinterlaced output. Obviously I re-interlace for DVD, but I keep the lossless deinterlaced restored version as an "archive" master, along with the original capture.


Hope this helps.


Cheers,
David.

[sanlyn]

Thanks for the input.

1. DirectShowSource isn't the best choice for a DV source. You should use AVIsource and make sure Cedocida is the DV codec used (typically by it being the highest or only DV codec in the video codecs list in Windows).

Did it just now. It's apparently the only DV codec AviSource will recognize here, because AviSource wouldn't work previously with DV's in my setups. It looks the same to me, but I'll keep it anyway.

The Sample 7 capture is far too bright. No crushed darks at least, but brights are blown away to smithereens.

They're not though, are they? There's absolutely no clipping. They're above 235 but below 255. Given that you're going to fix levels and colour correct, the fact they're above 235 is irrelevant. Some would even claim the extra range gives you more precision in the correction (I wouldn't - the amount of noise makes it irrelevant).

When viewed or played as RGB, green and Blue are beyond RGB 255. If a color smacks against the right-hand side of a histogram and climbs halfway up the right-hand wall, it exceeds 255. I realize that it's not absolutely necessary to go into RGB for processing. But I don't know how you can avoid displaying Rec601 in RGB on a monitor or TV. Nothing that I own displays a YUV color space.

Color that exceeds RGB 255 doesn't offer more precision. Clipped color has no dynamic range beyond the clipping point. All one could work with would be whatever remains below the clipping point -- in which case, colors like bright white might not even be possible to recreate because the "brights" in one or more colors would no longer exist.

One could avoid clipping some of that range extension by converting to RGB with the PC.601 matrix. I've always guessed that DV is for PC display and thus uses a similar matrix. But broadcast and non-DV such as MPEG don't use that matrix.

The image below is the original frame 119, unprocessed except for decoding with the codec suggested. The front of the bride's dress has very little detail, nor does her bouquet nor the flower in the groom's lapel. It's possible that these details got washed out in-camera. The image below frame 119 is original frame 649, which is darker lighting and shows far more detail. The top image has a fog or 'bloom" around the brights, to the point where bright green blooms against the bride's skin. I managed to fix that in Photoshop Pro with color channel level controls, but most users don't have such features available.


[Attachment 22808 - Click to enlarge]

[Attachment 22809 - Click to enlarge]

The color problems were just plain weird. In the top photo the groom's hair is a grayish green; in the bottom photo it's closer to yellowish but still too green. Another oddity is that red flesh tones in the bottom photo are too red, as well as looking desaturated. In fact color looks desaturated everywhere, despite the RGB extension of the brights). Raising saturation (I did raise it about 5% with ColorMill) blew red out of proportion to green and blue. I struggled against that effect for quite some time, but gave up on it; it would involvce reshaping the histogram in great detail. Skin tones were a real problem. In the top photo the bride's chest and shoulder have an unattractive grainy green mottling. There are filters for that (I tried camcorder color denoise, but it left other artifacts that were just as bad -- yet another filter that often exhibits strange behavior).

2. It's amazing that you've made white things white again, and got the dresses right, but I think the skin colours have suffered a little. I'm not looking on a calibrated monitor, and I'm not about to burn it to DVD to check, but I think the pinky skin colours of the original version may be more true-to-life than the browny ones of the corrected version. Something in between is probably correct.

I tried to correct for whites, grays, and skin tones. That didn't always work: it looks OK in some frames, not-OK in others. As I mentioned, the colors and levels keep changing between scenes and even during scenes. I used the same corrections for the entire video, but look at the consequences: As the photographer plays trombone with the zoom lens, detail in the church interior appears and disappears. In the second shot with the girls and the child entering the scene, the levels change from bright to dark and then red takes over in the latter part of the shot and the bride and everyone else look strange. If one corrects the last part that scene for that segment alone, then in the first part of the scene the girls are orange. The "common" settings make the last shot of the parents look too bright. This is a problem with VHS anyway: no two consecutive scenes will look alike, so you end up working with each scene. The gray suits were another glitch: I don't think they're really "gray", but some off hue toward blue, although they turn yellowish as the camera moves around.

If you consider tape aging effects, things get even more bizarre at times.

Everything seems to lack a certain element. It lacks a certain je ne sais quoi, pardon my French, or an element of depth or solidity. It looks painted or colorized-- I'm talking about the original here, not the output. Earlier I put some of these vids in After Effects with its Color Finesse and other sophisticated controls and layer work (still working with those damn layers, which I'll likely never get the hang of), and aventually got better results. But not without a struggle. That scene-by-scene tedium isn't something I'd think tafflad would go for.

3. I wish I could figure out a way of maintaining some noise on this.

So do I. I also wish I could maintain fine detail and well-defined edges. The combination of aliasing and really horrible edge ghosting really kills anything else one might want to accomplish. I used other means to smooth the aliasing at times, and different resizers. But the edge artifacts simply wouldn't let go. I scoured Doom9 for hours, and actually have spent days over time looking for fixes. There are routines with Mask Tools that address it, but none of them worked here. I have a feeling that MaskTools would likely offer something better, but I just don't know the plugin that well.

4. If I've deinterlaced something while restoring, I keep the deinterlaced output. Obviously I re-interlace for DVD, but I keep the lossless deinterlaced restored version as an "archive" master, along with the original capture.

Hope this helps.

It does. Your input is always illuminating. Thanks again.

[manono]

The combination of aliasing and really horrible edge ghosting really kills anything else one might want to accomplish.

That's what jumped out at me. It sometimes has two layers of edge enhancement, the usual lighter colored one and a second inner black layer. Plus the aliasing. I don't think it'll all disappear if he turns off the sharpness when capping, but he was encouraged to keep it on (not by me), and that part of the capture can't help but be improved by turning off the sharpener.

And, as 2Bdecided mentioned, the whites aren't so bad that they can't be easily fixed. And I see no signs of any clipped colors. I'm not quite sure how you're seeing these things. I cropped it to remove the black borders and :

AVISource("Sample 7 - Mid+Edit On+DVDR.avi")
Yadif(Mode=1)
Crop(40,0,-40,0)
Histogram(Mode="Levels")

Is there something wrong with that method? I don't see any chroma levels going much above 130 when removing the histogram and adding:

ColorYUV(Analyze=True).Limiter(Show="Luma")

Or much black/white beyond 16/235, for that matter.

[sanlyn]

And, as 2Bdecided mentioned, the whites aren't so bad that they can't be easily fixed. And I see no signs of any clipped colors. I'm not quite sure how you're seeing these things. I cropped it to remove the black borders and :

AVISource("Sample 7 - Mid+Edit On+DVDR.avi")
Yadif(Mode=1)
Crop(40,0,-40,0)
Histogram(Mode="Levels")

Is there something wrong with that method? I don't see any chroma levels going much above 130 when removing the histogram and adding:

ColorYUV(Analyze=True).Limiter(Show="Luma")

Or much black/white beyond 16/235, for that matter.

Nothing "wrong" with it. But IMO it shows how YUV data is stored, not as it displays on a '601' RGB device or when trying to work in RGB.

Below are two RGB histograms for frame the source frame 119 posted earlier. The histograms are from the unprocessed image. Right image=Rec601, Left image=PC601. You can see that PC601 stays within prescribed boundaries -- for PC's. But I wanted to correct for DVD/PC.

[Attachment 22817 - Click to enlarge]

The opening frames aren't too bad (aside from the terrible color), but later frames get a little haywire. The image and histogram below are the original frame 499. It looks washed out (because it is), but why so bright for subdued overcast or indirect lighting? There must have been a little weather that day, because few of the images appear to be in sunlight. Below, green and blue are out of bounds or near it. That's demonstrated by the lack of bright detail in the gowns. Green has crept past the right boundary of the graph.

[Attachment 22818 - Click to enlarge]

The vectorscope below shows how RGB display is affected. The green band is clipped against the upper edge of its channel, indicating that there might be some detail "overshoot" that could be recovered. Note the lack of detail in the brightest parts of the green gown of the girl in front.

[Attachment 22819 - Click to enlarge]

Below, a simple correction of ColorYUV(off_y=-10) makes things look a little more normal. But I still have a problem with the brightest greens.

[Attachment 22820 - Click to enlarge]

Below, more cotrrections, reducing off_y 2 more points, and gamma=-20 and contrast_y=-10. Still needs a little work, as the objects in the dark church are now obscured and so is some detail in the boy's black suit -- which looks green here because red is still depressed. More corrections followed, especially with taming that weird green and applying a contrast mask for the darkest areas. More work ahead, but this was the basic method I used.

[Attachment 22821 - Click to enlarge]

The point IMO is that luma might have been OK, but chroma for the intended display was not. This wasn'tr a simple matter of adding red contrast, but of controlling green. How green got "stretched" that way is a mystery to me, but it's in almost all of the videos the O.P. has posted. If it's not green, it's blue or red.

[magikarp99]

What software are you using to generate those histograms?

[davexnet]

I agree about the colorized look. Was there enough bandwidth in NTSC VHS to get more accurate/subtle color?
That's pretty much the way it used to look, in my opinion.

[sanlyn]

What software are you using to generate those histograms?

The histogram control itself displays as histogram, vectorscope, or waveform monitor. It's the Trevlac Color Tools plugin for VirtualDub. website here: http://trevlac.us/colorCorrection/colorTools.html -- which doesn't tell you all that much and has been around for years, but it shows an image of the VDub interface. The .vdf plugin is here: ColorTools.vdf (right-click and "Save As", depending on your browser). If you just left-click the .vdf link, all you get is a garbled display of the binary code. The download has existed since The Flood when introduced on Doom9 but never updated. Don't know why. Lots of people use it. Also documented on a couple of pages in Doom9's video processing guides somewhere and on digitalfaq.com, but it's easy enough to figure out on your own.

Doom9 has images showing its use as a waveform monitor and vectorscope at Postprocessing video using VirtualDub. Scroll down past the middle of the page to the title "7.1.9 Color adjustment". You'll see it mentioned in several guides and forums.

The images were captured from VirtualDub and resized and composed in Photoshop.

[magikarp99]

...

[magikarp99]

It's the Trevlac Color Tools plugin for VirtualDub.

Thanks

[2Bdecided]

sanlyn, I think you've written two very long posts arguing about a very simple point that no one doubts: I know the captured levels are wrong to display as-is. The point is that they don't exceed the built-in headroom of YUV. You're going to fiddle with them anyway, therefore the fact that they're wrong to display as-is is not a problem. You will bring them into range before they are displayed.

You should always take care if you need to convert from YUV to RGB, even if you think the levels are correct for display, and especially when you know they might not be.

Regarding the "PC.601" matrix in AVIsynth - for years I'd assumed it converted to StudioRGB (as some people call it), but in fact it does something different...
http://forum.doom9.org/showthread.php?t=164981
...there's a solution suggested right at the end of that thread, but I haven't tried it.


Cheers,
David.

[sanlyn]

Thanks again for finding links to posts that I lost but never could find again. As I saved the linked pages and then saved the pages they link to, I found that I had downloaded some of a while ago (but I saved with titles that threw me off track for finding those posts again!).

The discussion does bring up a point I have against VHS->DV capturing (i.e., color accuracy problems, etc.), but the point here is that in this particular case the YUV original did include some headroom. That YV12->RGB conversion problem is something I noted a long time ago. I used to post scripts that did some odd things, such as using a special YV12toRGB dll, or converting to YV12->YUY2->RGB first, or using the dither plugin to cross between YUV -> RGB and back. These little tricks just confused and confounded people, who ignored them.

I'm sorry to say that when I read tech stuff like this (which you have to do if you really want to get into processing), I understand what's happening but I don't remember the details. What I remember is the main point: you have errors or problems if you do so-and-so, but you can avoid that by doing such-and-such. Most of the time, if you post those methods you end up arguing about it and my lack of "detail" forces me out of the debate.

Anyway, thanks for the refresher. It helped me to find old posts I'd saved from trevlac and the like -- in this case, I first saved those posts and a dll back in early 2011. Thank goodness for big eternal hard drives, but I think I need to keep an updated guide on where the hell I save stuff and how I named them.

[davexnet]

I took the image posted earlier and tried to see if I could make it look more lifelike in Photoshop.
The most natural look seemed to be when trying to set colors that should be neutral to neutral.
The main problem is still the limited color bandwidth. They look a little warmer, whether it's any
better is debatable.

[2Bdecided]

The discussion does bring up a point I have against VHS->DV capturing (i.e., color accuracy problems, etc.)

The ADVC110 (at least) does a very simple repeatable thing.

PAL/625 voltages relative to blanking quoted:


It assumes the sync pulse should be at -0.3V.
It scales the entire video signal to make the sync pulse -0.3V.
It then digitises the luma with 0V mapped to 16 and 0.7V mapped to 235.


It assumes the colour burst should be 0.3Vpp
It scales the chroma to make the colour burst 0.3Vpp.
It then digitises the chroma with what appear to be the correct mappings (though off the top of my head I don't know what they are).


So, in fact, the ADVC110 exhibits the opposite of "colour accuracy problems" - it captures the video signal with near perfect colour accuracy.


Any faults in the source are also perfectly and repeatabley captured. That's not what you want of course - you want to be able to fix incorrect levels, and the ADVC110 gives you no way to do this. If they're wrong but not clipped, you can fix it in software. If they're wrong and clipped, you're stuffed.


But as long as it doesn't clip, it is a very accurate capture of the levels in the original video signal - it's just that those original levels themselves might not be right, accurate, or even consistent.




It's not a property of all DV capture devices. My HV20 in analogue>DV mode twiddles with the levels and reduces the noise - the results can be a big improvement (composite input only unfortunately), but are sometimes nothing like the original video signal.


Cheers,
David.