What are the best filters for these VHS captures?

[jwhittin]

Hi,

Firstly I'd like to thank everybody, particularly lordsmurf and sanlyn, for the wealth of useful information on this site. With your help I've already done a lot of the job of transferring my old family VHS tapes to digital just by reading your guides and posts. However now I've got all captured AVI masters for all these tapes I'm now somewhat stumped as to how to get the best (within reason) result from what I've got with Avisynth/VirtualDub etc.

I guess I should start with some background. I'm in Australia, so we're PAL.

Whilst I totally accept your reasons for specific JVC and Panasonic semi-pro VCRs and TBCs, the cost to get them and ship them to Australia (even if I could resell them when finished) wasn't justified for the number of videos I had. So I've had to make do with what I had - a Sharp VC-H660 and a borrowed LG CC993TW. The LG seemed to produce slightly better results so that's what I've used for all the tapes.

For capturing, I already had a DigitalNow DNTV Live! Dual Hybrid PCIe which was, for years, used in my Media Centre. I believe it uses a Philips (NXP?) SAA7162 chip. Input was unfortunately by composite as, due to my decision re VCRs, neither VCR has S-Video outputs.

This is where it gets interesting. If I used VirtualDub (1.9.11 as per your recommendations) I could never get lip sync right - it would progressively get worse over the tape, even if audio and video display were disabled, network cable was disconnected, etc, etc. My guess is that it had something to do with dropping frames. After lots of failed captures using different settings and lots of reading of this site I finally tried iuVCR and iuVCS and they nailed it first time - no fuss or fiddling. So that information may be useful to others.

So I now have AVI masters captured using iuVCS which generally (with one exception that I'll refer to later) I'm fairly happy with, particularly as they are by composite with no TBCs. They are/will be archived off onto 1TB drives which were otherwise unused (my server outgrew them!).

The main issue I have is that the resultant images often have jagged edges - even after spending a lot of time on this site I'm not sure of the correct term. I'm presuming it's something to do with interlacing but I'm not sure. It is most pronounced when the camera itself pans. There may be some other issues, but it's this issue which is my main concern.

What I want to do now is create versions for normal viewing without these jagged edges (or any other faults easily fixed). Most of our viewing is likely to be using Windows Media Centres (or XBox 360 extenders), for which DVR-MS (ie MPG) works best but H264 MP4 works OK, but a limited amount of viewing may be on notebook PCs or tablets. So perhaps a relatively high bit rate DVR-MS/MPG version for normal use which I can then further convert if necessary to H264 MP4 if needed? I'm very open to suggestions.

I've attached two videos - Wedding Outdoor and Wedding Indoor - to give you an idea of what I've captured. Most of my other videos are fairly similar - it's just that our wedding is the most important one to get right.

The trouble is now, and the one that I'm not solving by further reading of posts on this site, is what settings/script I need to use on Avisynth (which I've got set up and tested) to solve the jagged edge issue and any other problem on the captures. Can someone help?

My second, less pressing issue, regards two videos which were taken (from our honeymoon) on a tour we did in Rio. They were recorded by the tour company and given to us as PAL. But the colour is very wrong - it was like that on the VHS so not an issue of a capture problem. I'm aware that Brazil has a different version of PAL so perhaps that's the problem. I've attached two samples from these videos - Rio 1 and Rio 2. Is there any way of making these a bit more realistic? What settings/script in VirtualDub or Avisynth would help?

As an aside I got a professional transfer done of this same wedding video when I was getting nowhere on lip sync. He used a professional VCR and TBCs but his version, other than not having the jagged edges, seems worse to me than my own captures - the aspect ratio is wrong which is easily fixed, and it seems grossly over processed (sorry I can't use the correct term for this but it seems to have lost a lot of dynamic range, eg less detail in faces) which is not easily fixed. And this is from one of the best reviewed converters in our state.

Many thanks for all your help so far, and even more in anticipation of some help re settings/scripts to fix the challenges remaining.

John

[themaster1]

For the record , as i'm currently looking for a good multistandard TBC myself, this company is based in UK but also New zealand (so close to you) and appears to have good gear (unverified claim). This one for example.

Your Rio videos have washout colors and the levels must be fixed, Not much jitter. Haven't check the others.

Forget virtualdub (even though it's handy) Learn avisynth scripts, that's the real deal for enlightened video amateurs (you can load vdub filters in avisynth scripts aswell for the most part)

But to answer your question: gradation curves, camcorder color denoise, cmyk, colormill, rgb equalizer 2.2 and of course neat video (latest version, not free). These are the best i mess with usually

[lordsmurf]

Quote:

Originally Posted by themaster1

For the record , as i'm currently looking for a good multistandard TBC myself, this company is based in UK but also New zealand (so close to you) and appears to have good gear (unverified claim). This one for example.

As mentioned in your VH thread, this will not work as TBC.

It's a conversion TBC only:
- PAL > NTSC = works
- NTSC > PAL = works
- PAL > PAL = does not work
- NTSC > NTSC = does not work

Not changing the signal throws the unit into a bypass mode, so no TBC.

I'll revisit this thread again soon, to answer all questions, but wanted to mention that immediately. Don't want the OP (or you) to buy the wrong item for $$$. It won't do as you want/need. Not the purpose of that device.

[jwhittin]

Many thanks lordsmurf! Very much appreciated.

[sanlyn]

Quote:

Originally Posted by jwhittin

This is where it gets interesting. If I used VirtualDub (1.9.11 as per your recommendations) I could never get lip sync right - it would progressively get worse over the tape, even if audio and video display were disabled, network cable was disconnected, etc, etc. My guess is that it had something to do with dropping frames

It could, but more likely it has to do with synergy between system components, which in itself can result in dropped frames. I'm absolutely useless with questions about audio sync and VirtualDub. I've never had that problem. Been capturing with AIW cards and VirtualDub since Windows 95 and have yet to encounter audio sync issues. I've used Virtualdub versions and the same AIW product line with 4 different cards and 7 different PC's and two non-ATI USB capture cards, and still no sync problems. I even tried to induce sync problems by removing all tbc elements but with no success, except for starting a capture with the F5 key instead of the F6 key (F5 produced audio slowdown after about 15 minutes of tape). Otherwise I have no answer.

Quote:

Originally Posted by jwhittin

The main issue I have is that the resultant images often have jagged edges - even after spending a lot of time on this site I'm not sure of the correct term. I'm presuming it's something to do with interlacing but I'm not sure. It is most pronounced when the camera itself pans. There may be some other issues, but it's this issue which is my main concern.

That's a fault of most consumer cameras and their shutter design. The only real fix is to double-rate deinterlace with medium-strength QTGMC and apply a mild santiag Avisynth filter (which will cost you some line detail), then leave the video deinterlaced at 50fps. But that's the least of your problems. The worst problems can't be fixed. Besides poor detail playback and some posterizing effects, there is severe tracking distortion and line twitter. This is especially noticeable in the Rio2.avi if you will look at the distortion and flutter in the upper left quadrant. Try playing this video frame by frame in VirtualDub with the yadif double-rate deinterlacer turned on (keep all frames, top frame first) and the distortion will be very clear. It also looks as if fields are being dropped, not frames. The outdoor wedding shot, which has very poor detail, also shows line twitter and "flashing" in the church brickwork when played as frame by frame with deinterlacing turned on so that you can see how detail varies and objects shift frame by frame while the camera zooms. These are tracking problems that can't be fixed with filters.

[jwhittin]

Thanks sanlyn and sorry about the delay - it's been a busy (long) weekend here and I needed some quality time to better read through, understand, and test your answer.

My comments about the capture lipsync were so you would know the origin of the capture but mainly for info so perhaps someone in future with the same problem and the same chipset will see my comments and may be able to save some time by trying iuVCS/iuVCR straight away.

I've viewed the samples as you've suggested and see the issues you've mentioned. Just out of interest, are these problems you mention mainly caused by the camera/recording or my non ideal playback device?

I'd like to get as good a result from the wedding video as I can without spending ~EUR1000 plus freight to Australia plus risk of unit failure to get a recommended one. So if you think I can get better capture from another consumer unit, I can try experimenting with others from friends and family. Othewise I'll just have to live with what I have and do the best I can post capture. At least I have three separate copies of the video so I can compare the results from each.

I never had high expectations for the Rio video as it was one of those "we'll give you a video of your tour you were on". It's just that it would be nicer if I could fix the colour so it's slightly less bad.

Given your filtering suggestions, how is this for a script (based on lordsmurf's MultiScript and I'd use different cropping for the Rio videos):

AVISource("Wedding Indoor.avi")
ConvertToYV12 # script below this line requires YV12
### Deinterlacers
QTGMC(Preset="Slow") # best deinterlacer - balances speed + quality
SelectEven() #
### Antialiasing + compensation for deinterlace
santiag(strh=2,strv=2) # antialiasing
### Typical VHS overscan crop
Crop(12,0,-12,-20) # use both lines
AddBorders(12, 10, 12, 10) # use both lines

This leaves a few related questions:
1. Is there a way to sensibly move back to 25fps or am I now effectively stuck at 50fps?
2. I note that the script above produces a file of 295MB from a 36.5MB input. I can understand why it may be double the size (due 50 vs 25fps) but why the rest of the increase? Have I done something wrong? This suggests that doing this for the 2 hours 40 minutes of the wedding tape will mean my ~90GB original will blow out to over 720GB.
3. With the Rio videos, can I fix the lousy colour?

Finally, I see that lordsmurf prefers broadcast bitrate MPEG-2 for archiving. This suits me fine as I can have DVR-MS files without recoding. But what bitrate and which converter? I already use VideoReDo TV Suite V5 a lot so am very familiar with it. Would this be suitable or do I need to use something else? Or has the view changed since that post on 2011?

Many thanks again for your help.

John

[lordsmurf]

The Brazilian videos are converted badly between formats, likely using one of the cheapo analog-domain format converters. PAL-M is essentially NTSC with PAL color. Converting NTSC to PAL (or PAL to NTSC) with a cheap converters always results on severe linear artifacts and ghosting. Maybe 50% (at best) of the ghosting can be removed with Avisynth.

But the linear damage is usually permanent. No amount of anti-alias will fix it, even slightly. Sometimes you can run both hard anti-alias, with a moderate vertical blur, to make it less annoying to watch. But it will make the video softer as a result. So pick your poison: soft or distracting. I opt for soft, because I despise interlace jaggies.

Deinterlace jaggies are often easier to correct, compared to interlace linear errors from format converters.

The Rio video also looks stretched. Geometric problems also exist with the cheap converters.

Your clips had obvious timing wiggles, the problem with lack of TBC.

The video quality is luminous, something weird going on in the luma channels. Philips SAA chipset are infamous crap, and this is probably what's caused it. The SAA7113 chipset is the main reason that Easycap/EZcap is EZcrap, and this is a primary issue. I mention this because it limits what can be done in color correction. In essence, contrast is bad, with illegal brightness. When you screw with lighter chroma/luma, you skew colors. And what you have here is a video with fairly decent darks, but off-tint lights. And there's really no way to color correct it.

Composite didn't help with the color problems.

At very least, in Avisynth, you can realign the chroma,a nd then follow it up with CCD in VirtualDub. That will at least clean up most of the color bleeding.

Audio hiss seems mild, but correctable, be it freeware Audacity or payware Sound Forge.

The "professional" job you had done, if it removed the jaggies, probably opted for softer video. Because, again, that's really the only way to remove harsh jaggies from interlace format changes. Hence processing. But I'm just guessing, and would need to see that version to make more judgements.

Sync can be complex. VirtualDub probably will work, but needs tweaks. Not all cards like the default settings. iuVCR isn't better, just different. But VirtualDub should be able to mimic the settings.

[sanlyn]

Quote:

Originally Posted by jwhittin

I've viewed the samples as you've suggested and see the issues you've mentioned. Just out of interest, are these problems you mention mainly caused by the camera/recording or my non ideal playback device?

Caused by both.

Quote:

Originally Posted by jwhittin

I'd like to get as good a result from the wedding video as I can without spending ~EUR1000 plus freight to Australia plus risk of unit failure to get a recommended one. So if you think I can get better capture from another consumer unit, I can try experimenting with others from friends and family. Othewise I'll just have to live with what I have and do the best I can post capture.

That's about all that most of us can do. Even with better gear, VHS still has [problems-- just not so many.

Quote:

Originally Posted by jwhittin

I never had high expectations for the Rio video as it was one of those "we'll give you a video of your tour you were on". It's just that it would be nicer if I could fix the colour so it's slightly less bad.

The Rio samples are far better detailed and look less "processed" than the wedding shots. I don't think the Rio color looks that bad -- but at the moment I'm working with a Dell laptop until some home construction is finished. Dell makes some of the worse laptop screens in the world, they can't be calibrated accurately, and working with video on a laptop is the pits. I'd be able to advise better about color later. Meanwhile I can say that the Rio black levels and gamma are too high, making the samples look a bit washed out, and brights are clipped.

Quote:

Originally Posted by jwhittin

Given your filtering suggestions, how is this for a script (based on lordsmurf's MultiScript and I'd use different cropping for the Rio videos):

.
I wouldn't suggest a one-size-fits-all script for any video. With VHS I wouldn't even suggest it for every scene. In this case the script is inappropriate on several counts. First, QTGMC at "slow" destroys whatever small amount of detail is in the wedding shots and won't do anything for Rio. A faster preset would be better here. Second, the script does nothing to correct the levels problems, which should be corrected in the original YUV and usually can't be corrected later in RGB. Third, to address another question:

Quote:

Originally Posted by jwhittin

Is there a way to sensibly move back to 25fps or am I now effectively stuck at 50fps?

The script shown doesn't output 50fps. The SelectEven() statement throws away every other frame and discards 50% of the temporal resolution. The output is 25fps progressive. Even if the script didn't throw away frames, QTGMC as used here will result in jerky video because it is using the wrong field order, which Avisynth assumes is BFF. Your videos are TFF. Also, QTGMC shouldn't even be used to deinterlace your Outdoor Wedding shot: that sample has already been deinterlaced by discarding alternate frames.

This also struck me as not quite inappropriate:

### Typical VHS overscan crop
Crop(12,0,-12,-20) # use both lines
AddBorders(12, 10, 12, 10) # use both lines

That cropping is anything but typical. Borders will vary by tape and by capture gear. On my Panasonic AG-1980 and ATI 9600XT AIW card the side borders are rarely even on both sides, with usually the left side having a thicker border, and never more than 8 pixels of head switching noise along the bottom (and it looks as if a handful of original image pixels along the bottom are already cropped by the 1980). On my PV-S4670 and PV-8664 VCRs, the side borders look about the same but there are only 4 pixels of bottom noise, sometimes only 2 with some retail tapes. The cropping used in lordsmurf's script doesn't really hurt anything, but it does throw away some image real estate. I'd suggest something customized for the video samples:

Crop(8,0,-8,-12).AddBorders(8,6,8,6)

Quote:

Originally Posted by jwhittin

ConvertToYV12()

If you want to avoid chroma damage, always specify interlaced r not with colorspace conversions. Use ConvertToYV12(interlaced=true) or ConvertToYV12(interlaced=false). The same for other matrices such as YUY2 and RGB. After running a deinterlacer, interlaced would always be false.

[lordsmurf]

I don't think recording had any real effect. Maybe some, but not likely. I see nothing worse than normal. The main quality killer was the conversion, the playback, and the capture.

My script isn't "one size fits all", but rather "starting point for newbies for common problems". It's supposed to be the script created for a guide series on Avisynth.

I've gone over this before: QTGMC is an EDI based deinterlacer, meaning interpolation, so source frames are not "thrown away", only post-interpolated/processed frames (aka created frames) are 50% discarded to retain source fps. Whether motion is truly lost is debatable, since no true source frame existed (half data at full motion = arguably same as full data at half motion, aka the original idea behind interlacing, and why film is usually perceived as better than TV). There's no way to deinterlace losslessly. Furthermore, 50fps/59.94fps isn't compliant to anything but computer/streaming viewing. In an ideal world, we'd have progressive at 50/60fps (or more) with formats to support it at both SD and HD resolutions. So the choice of "throwing away" is not always a choice, but a requirement.

AssumeTFF() is in my current script.
Same for ConvertToYUY2/YV12(interlaced=true/false).
This was an omission in my first MultiScripts. I'm hoping to update that soon.

The overscan is typical for my JVC VCRs.

[sanlyn]

Quote:

Originally Posted by lordsmurf

I've gone over this before: QTGMC is an EDI based deinterlacer, meaning interpolation, so source frames are not "thrown away", only post-interpolated/processed frames (aka created frames) are 50% discarded to retain source fps. Whether motion is truly lost is debatable, since no true source frame existed (half data at full motion = arguably same as full data at half motion, aka the original idea behind interlacing, and why film is usually perceived as better than TV). There's no way to deinterlace losslessly. Furthermore, 50fps/59.94fps isn't compliant to anything but computer/streaming viewing. In an ideal world, we'd have progressive at 50/60fps (or more) with formats to support it at both SD and HD resolutions. So the choice of "throwing away" is not always a choice, but a requirement.

I guess we'll have to continue to disagree. I won't be taking any of my 800-plus DVD's and BluRays and downgrading them from 59.94 images per second to 29.97 images per second. That those ignored frames are interpolated versions of original fields is irrelevant to the fact that 29.97 is only one-half of 59.94. How anyone can insist that 29.97 and 59.94 are the "same thing" escapes me at the moment, as well as the notion that one interlaced field is the same thing as two interlaced fields, since the math that I learned keeps telling me that zero + 1 is not two. To say that QTGMC can't deintelace losslessly is not accurate, as its documentation explains a lossless mode. Further, you're wasting time interpolating a fake frame and then throwing it away with SelectEven(), when FPSDivisor can be used instead. By downgrading a video from 59.94 images per second to 29.97 images per second, you're kind of stuck with the argument that SD BluRay and 1920x1080 BluRay at 29.97fps progressive are definitely non-compliant and must be encoded and authored with fake interlace flags. This means that the video at hand is now forced into telling a player to deinterlace a fake-interlace progressive image, upscale the images yet again to two duplicate full sized frames or otherwise show the same frame twice during play. Having done a lot of programming over the years, I can definitely say that doing something twice isn't very efficient, and that it looks odd while it's being done 3600 times a minute. Also, showing the same image twice during play doesn't do anything to smooth the jerky motion (which some people can't see because, truly, they'll watch anything and can't tell one thing from another anyway). Another problem with reducing temporal resolution from 59.94 or 50 images per second to 29.97 or 25 images per second is problematic for computer viewing, when there are no 30 or 25fps consumer monitors out there. Nobody views a computer screen at those reftresh speeds.

So I guess anyone can watch half a video and live with it. I can't.

Film is usually considered better than TV for a multitude of reasons but they aren't related to throwing away interlaced fields. You can't play film at 23.97, 19.98, 17.98 or 15 fps on a modern TV, which would be incapable of generating an acceptable image. You have to start doing oddball things like inventing all kinds of telecine (duplicate framing and interlacing techniques) that clearly shows up on today's TV's as clumsily duplicated images shown several times in a row -- and as far as I know, repeating images during play used to be called stutter. I don't know what it's being called today. People already complain about it when it's done for 2 out of every 5 frames during playback of 3:2 pulldown. Producers vtry to get away with it with gear that has 23.97 and 24 fps playback "modes", but it really does look kinda funny during motion when the screen's refresh rate isn't anywhere near frame rate.

I also have to ask that if that "other" field doesn't exist and doesn't matter, why is it being captured in the first place? Should we advise users to capture only as progressive, to find a capture device that captures only the top field or bottom field and ignores the other? They actually do make capture cards that do that, but we advise people to avoid them because they're -- ahem -- dropping fields. I believe it's really called "dropping" isn't it? Don't people pay big money for Adobe Premiere Pro which can deinterlace by dropping fields? Isn't that a menu choice in Adobe and other NLE's, and doesn't VirtualDub offer the same choice? Why ask slowpoke QTGMC or pricey Adobe to do it, when it can be done during capture? After all, we're telling people that the ghostly and mouse teeth combed second "field" they see in their editors doesn't really exist. Why capture it?

I don't even want to get started on how incredibly awful a piece of digital video looks next to a film.

[lordsmurf]

It's just the typical quality exchanges we have to put up with for SD video.

- You can get half data at double motion, with linear artifacts (interlaced).
- Or you can get full data at half motion, with potential motion softness (film).
- Or something in between (deinterlaced), with the flaws of both, and never superior in quality to either.

And that last option can break compliance with established formats, such as retaining the 50/59.94 fps.

It is what it is.

Remember, I don't necessarily like any of this either. I've never been overly fond of interlace, and deinterlace has always been a boondoggle. I've seen vast improvements in the past 15+ years, with QTGMC the current champion. But it's complicated and misunderstood.

QTGMC has a setting called "lossless", but there's not a way to really have a lossless deinterlace. That'd be a bob, and bob is one of the oldest and lowest quality methods. Selur from Doom9/etc (the developer of Hybrid) probably understands this better than I do, and he's also stated this in a recent conversation. He's also very careful to explain the SelectOdd/Even come after interpolation.

QTGMC is not just an EDI, but does anti-aliasing, gaussian blur, and other processing. It's amazingly complex at recreating the video with new information.

My entire disagreement with you is the term "throwing away", as it implies source data is being destroyed. It simply discards interpolated data. Yes, each new frame has some source data, but also new data. Ideally, yes, of course, keep it if possible. But it also breaks compliance with every current format, meaning we often have no choice to discard (aside from Even or Odd).

23.976>29.97 is called judder, and it looks terrible. Many modern HDTVs and players play raw 24, and it fits nicely into the 120Hz/240Hz. Really smooth and fluid motion -- but of course some complain about it (the so-called "soap opera effect", which has nothing to do with soap operas). DVDs hard-core the judder 24>30, and many HDTVs also have dejudder filters, with varying success (example: Sony = excellent, Samsung = lousy).

A lot of HDTVs will also interpolate. So you regenerate that 50/59.94 double rate. It all artificial data anyway. What you say about not destroying and then reinventing is accurate, but ultimately not that drastic. Again, keep when you can (streaming), not catastrophic when it's not (DVD, BD, etc).

We're not really disagreeing on much. You just need to soften your stance some.

The most important thing is we're both advocating QTGMC, just differing on the exact settings choices.

Film vs. digital mostly depends on digital sensors and film stock. Both can look great or awful.

@jwhittin: For that MP4/streaming version, go ahead and do as sanlyn says, keep the full 50fps frame rate. Note that the lossless file will be 200% size, so about 70gb/hour (about the same as uncompressed YUY2 size), so be prepared for that.

[sanlyn]

Debates about discarding data aside, I didn't answer this question:

Quote:

Originally Posted by jwhittin

1. Is there a way to sensibly move back to 25fps or am I now effectively stuck at 50fps?

My position is that the script already has you at 25fps, since all odd-numbered frames are ignored. If you want format-compliant interlaced video, you keep all the deinterlaced frames. To do that, you remove the SelectEven() statement and change this from the original script:

QTGMC(Preset="Slow") # best deinterlacer - balances speed + quality
SelectEven() #
### Antialiasing + compensation for deinterlace
santiag(strh=2,strv=2) # antialiasing
### Typical VHS overscan crop
Crop(12,0,-12,-20) # use both lines
AddBorders(12, 10, 12, 10)

Change that code to retain all the frames and reinterlace at the end:

change to this re-interlace version.

QTGMC(Preset="Slow")
### -- filter...
### ---filter...
### ---etc....
### Crop and restore borders
AssumeTFF()
SeparateFields().SelectEvery(4,0,3).Weave()  ### <-- re-interlace to 25i

I can try to tackle color issues when I'm back on a proper monitor shortly.

[sanlyn]

I may have neglected to thank you for posting your avi samples, so this is to acknowledge your contyribution. You'd be surpirsed that it's sometimes like pulling hens' teeth to get samples posted.

Quote:

Originally Posted by jwhittin

The main issue I have is that the resultant images often have jagged edges - even after spending a lot of time on this site I'm not sure of the correct term. I'm presuming it's something to do with interlacing but I'm not sure. It is most pronounced when the camera itself pans. There may be some other issues, but it's this issue which is my main concern.

The term is aliasing, but some of it line twitter (similar). You will find them in non-interlaced video as well. They are due moatly to shutter designs of the day, but they still occur in consumer digital cameras. The effect is a high-drequency disturbance.

Two methods are used. First, the lowpass method: Upscale the image to 2x or larger to make lines thicker, then apply anti-alias smoothing and blurring. Then downsacle to 1/2 size to combine some of the smallest dots-and-curls in the lines. Then normal resize. The second method is shutter blur technques, sometimes coupled with low-pass. Both metrhods soften the video, with shutter blur usually the softest. Neither method is a 100% fix.Interlace makes it look worse, because each interlaced field is affected in a different way, so you have one frame's problem combined with the other's.

If I used VirtualDub (1.9.11 as per your recommendations) I could never get lip sync right - it would progressively get worse over the tape, even if audio and video display were disabled, network cable was disconnected, etc, etc. My guess is that it had something to do with dropping frames.

Quote:

Originally Posted by jwhittin

If I used VirtualDub (1.9.11 as per your recommendations) I could never get lip sync right - it would progressively get worse over the tape, even if audio and video display were disabled, network cable was disconnected, etc, etc. My guess is that it had something to do with dropping frames.

Dropped frames indicates a missing frame-level tbc. It also suggests that you set up ViurtualDub defaults to drop frames on its own. I don't know if you've seen this later guide to VDub capture settings: Capturing with VirtualDub [Settings Guide].

Quote:

Originally Posted by jwhittin

As an aside I got a professional transfer done of this same wedding video when I was getting nowhere on lip sync. He used a professional VCR and TBCs but his version, other than not having the jagged edges, seems worse to me than my own captures - the aspect ratio is wrong which is easily fixed, and it seems grossly over processed (sorry I can't use the correct term for this but it seems to have lost a lot of dynamic range, eg less detail in faces) which is not easily fixed.

Both wedding samples appear over-filtered, but in different ways. The indoor shot has been blurred to mask aliasing, apparently, and both shutter blur and low-pass blurring seem to be used. The outdoor shot is over-sharpened, not to mention overly-denoised, and appears to have been deinterlaced by dropping alternate frames) to mask some of the line twitter in the church stonework. Both appear to be played with a VCR's noise reduction in turbocharge mode.

I'm posting notes and suggested scripts on all 4 of the samples, with the Rio avi's first. I am also posting the results of the scripts as lossless avi's compressed with lossless Lagarith as YV12, ready for encoding as you desire (mp4, BluRay, DVD, etc. are encoded as YV12). I've used Lagarith because huffyuv can't compress YV12. Lagarith is huff's main competitor and is easily installed with own installer, which installs both 32-bit and/or 64-bit versions, as needed for your OS. For more info on Lagarith, the home page is here: https://lags.leetcode.net/codec.html. The Lagarith installer is posted in the forum here: LagarithSetup_1327.zip.

Rio 1.avi

The color is hampered with a little red over saturaation, but the color balance is normal. It's a lesson in what you're camera is doing, which is washing out colors. You have a dark foreground and a bright background, and the camera is trying to balance the two by lightening the foreground. This sets the black level too high and kills contrast. In the code I use a y-channel offset (y=luminance or brightness) to reduce the brightness of every pixel by a specified amount, lowering the blackest blacks to y=16. In RGB display, y=16 will be "stretched" to RGB = zero black. There is also some overrun in the bright spectrum, so some clipping. I used code to maintain luma levels in the 16-to-235 safety range.

There is also a bad dropout that I repaired with ReplaceFramesMC to interpolate new frames. The dropout is a bad ripple across the middle of the screen. It begins in the bottom field of frame 91 and takes up both fields in frame 92. The frame numbers shown in the code are deinterlaced numbers. The flicker you see in the video is due to each video head playing a different and slightly distorted version of each field. This also results in a kind "blinking noise" throughout the video. The only fix for this is a VCR with better tracking.

I've attached a copy of the ReplaceFramesMC filter. I'm also attaching a hard to find plugin named LimitedSharpenFaster (LSFmod is the newer version, but the older "LSF": is still the favorite). Both filters are supplied as .avs scripts and belong in your Avisynth plugins folder. Because they are .avs files instead of dll's, they must be imported explicitly into a script using the Import() command, which is illustrated in the code below. Why avs filters? Often there are several versions of the same filter with common internal coding; scripted filters can reduce version conflicts because avs files aren't loaded automatically like dll's.

### --- Edit the path to match location of plugins in your sytsem ---#
Import("Drive:\Path\To\AVisynth\plugins\ReplaceFramesMC.avs")
Import("Drive:\Path\To\AVisynth\plugins\LimitedSharpenFaster.avs")

### --- Edit the path to match location of video in your sytsem ---#
AviSource(vidpath+"Rio 1.avi")
ColorYUV(Off_Y=-12)
Tweak(cont=1.1, sat=1.5,dither=true,coring=false)
Levels(16,1.1,255,16,235,dither=true,coring=false)
AssumeTFF()

ConvertToYV12(interlaced=true)
QTGMC(Preset="medium",sharpness=0.7,border=true)
vInverse2()                #<- reduce exessive interlace combing                        
Santiag(2,2) 

### --- repair bad frames, numbers are deinterlaced frame numbers ---#
ReplaceFramesMC(183,3)
Crop(12,0,-4,-16).AddBorders(8,8,8,8)


###--- Re-interlace for DVD, BluRay ---###
SeparateFields().SelectEvery(4,0,3).Weave()
return last

Rio 2.avi

This is the frequent aliasing and line twitter on diagonals during motion that gives me fits as well as giving them to you. I've had this same problem with NTSC VHS and my sister's kamikaze camera antics. The original seems slightly softened anyway, and low-pass I used in the script does a little more of it, but it seems tolerable. There is distortion in the upper left quadrant of the image, and the video has non-safe luma levels that exceed y=16-235. The sky has no blue because the brightest blues were destroyed by bright clipping.Again, black levels were too high.

The line twitter is so aggressive here that the only way to subdue it entirely (almost) is to discard alternate frames after deinterlacing. This makes for a 25fps progressive video. The slight choppiness in motion is difficult to notice. For DVD or BluRay, the video can still be encoded with what are called "fake" interlace flags. It's not illegal, since most players will play the video as interlaced anyway. Meanwhile the color balance was good, so lowering black levels and raising saturation bring the colors out and generate some natural contrast to give the image dimension. The overall color balance was originally slightly blueish. I also used the VirtualDub DeFlicker flicker to soothe some of the luma filter, notably seen across the top and right side of the image in the original avi.

### --- Edit the path to match location of plugins in your sytsem ---#
Import("Drive:\Path\AVisynth\plugins\LimitedSharpenFaster.avs")

### --- Edit the path to match location of video in your sytsem ---#
AviSource("Drive:\path\to\Rio 2.avi")

ColorYUV(Off_Y=-12,off_u=-3,cont_v=-35)
Tweak(cont=1.1, sat=2.5, dither=true, coring=false)
Levels(16,1.1,255,16,235, dither=true, coring=false)
AssumeTFF()
ConvertToYV12(interlaced=true)

###--- QTGMC discards alternate frames with FPSdivisor for 25fps output ---#
QTGMC(Preset="medium",FPSdivisor=2,sharpness=0.7,border=true)
Santiag(2,2)
LimitedSharpenFaster()
Crop(8,0,-4,-16).AddBorders(6,8,6,8)
return last

Wedding Indoor.avi

Some shutter blur or other blurring effect has been added, probably to mask serious aliasing and twitter with all the motion. But it's overdone, and the source video doesn't seem to have that much detail to begin with deu to built-in noise reductiopn. It's because of effects like these that many users turn off noise reduction or use a VCR that doesn't have it. The color balance is good, maybe a little under saturated. After deinterlacing I applied the Unfilter() plugin, which attempts to undo previous blurring. Use this filter carefully, as it actually tends to over-sharpen! The object here was to upscale by 2X to make edges thicker, unsharpen, then sharpen, then downscale to normal size and sharpen again.

### --- Edit the path to match location of plugins in your sytsem ---#
Import("Drive:\Path\AVisynth\plugins\LimitedSharpenFaster.avs")

### --- Edit the path to match location of video in your sytsem ---#
AviSource("Drive:\path\to\Wedding Indoor.avi")

ColorYUV(OFF_Y=5)
Tweak(sat=1.25,cont=1.1,dither=true,coring=false)
Levels(16,1.1,255,16,235,dither=true,coring=false)
ConvertToYV12()
AssumeTFF()

### --- Temoporary crop for Histogram tests ---###
#Crop(10,0,-10,-16)
#ColorYUV(Analyze=true)
#Histogram("Levels")
#return last
### --- END of Histogram tests ---###

QTGMC(Preset="super fast",border=true) #,tr2=3,Sharpness=0.7,SLMode=2,border=true)
nnedi3_rpow2(opt=2,rfactor=2,cshift="spline36resize").UnFilter(100,100).LimitedSharpenFaster()
Spline36ReSize(width/2,height/2)
LSFmod(strength=250)

Crop(10,0,-8,-16)     # Crop off sloppy, uneven borders
AddBorders(8,8,10,8)  # restore original frame size
###--- Re-interlace for DVD, BluRay ---#
AssumeTFF().SeparateFields().SelectEvery(4,0,3).Weave()
return last

Wedding Outdoor.avi

This was definitely over filtered and appears to be played by a vcr whose dnoising destroyed much fine detail and oversharpened as well. The person processing also decided to deinterlace by discarding alternate frames, so this .avi sample is already deinterlaced but the line twitter between church bricks persists (that those last few words would make quite a tongue twister!). I used blurring methods here, not so harmful since the image is already oversharpened. Again, for DVD or BluRay this would be encoded with fake interlace flags.

The Avisynth filter "RemoveDirtMC.avs" is attached. Place it in AVisynth plugins. There are 9-plus versions of RemoveDirt.

### --- Edit the path to match location of plugins in your sytsem ---#
Import"Drive:\Path\AVisynth\plugins\RemoveDirtMC.avs")
Import("Drive:\Path\AVisynth\plugins\LimitedSharpenFaster.avs")

### --- Edit the path to match location of video in your sytsem ---#
AviSource(vidpath+"Wedding Outdoor.avi")
cOLORyuv(off_v=2,7)
Tweak(sat=1.75,dither=true,coring=false)
Levels(16,1.1,255,16,235,dither=true,coring=false)
AssumeTFF()

ConvertToYV12(interlaced=true)
QTGMC(InputType=2,Preset="medium",tr2=3,Sharpness=0.7,SLMode=2,border=true)
RemoveDirtMC(40,false)
ConvertToYUY2(interlaced=false)
nnedi3_rpow2(opt=2,rfactor=2,cshift="spline36resize").Blur(0.3,0.5).Blur(0.3,0.3)
SPline36Resize(352,288)
SPline36Resize(720,576)

ConvertToYV12(interlaced=false)
LimitedSharpenFaster(strength=75,edgemode=2)
AddGrainC(1.5,1.5)
Crop(12, 0,-8,-14)     # Crop off sloppy, uneven borders
AddBorders(10,6,10,8)  # restore original frame size
return last

[jwhittin]

Wow, that's very detailed and helpful information. Many many thanks to you both.

Sorry for the delay - I've been offline for two days for various reasons. But I'm now back online and will need some time to digest and test your advice.

In the meantime I've got a few clarifying questions to ask:

All the samples sent were direct from the iuVCS capture, ie no processing by me or anyone else after taking off the PAL VHS tape. So I'm a bit confused by the suggestion at least one is already de-interlaced. The wedding videos were captured in 1999 on "digital" by a semi-pro so I suspect it may have been a DV camera with the results copied onto VHS (I've got three copies). Surely he wouldn't have de-interlaced before saving it to VHS? We never got a copy of the DV and have lost contact with the guy who did the video so I think we can assume that the VHS is the best we'll get. The Rio videos came from the tour company, so I suspect LS's suggestions are pretty much spot on.

The professional transfer that I mentioned as an aside only produced one (incorrect aspect ratio) MP4 of the wedding video and has not been uploaded. Sorry if that comment caused any confusion.

sanlyn suggests that the VCR could be too agressive. They were all played on a LG CC993TW (6 head) but I've also got a Sharp VC-H660 (4 head) sitting right next to it. Is it worth posting one of the same AVI clips from that to see if that is a better source? To my untrained eye the LG seemed better - hence using it - but perhaps it's not...

Quote:

Dropped frames indicates a missing frame-level tbc. It also suggests that you set up ViurtualDub defaults to drop frames on its own. I don't know if you've seen this later guide to VDub capture settings: Capturing with VirtualDub [Settings Guide].

I tried quite a lot of various settings of dropping/keeping frames but could get none to work. There is a probably a setting which works but after a week or so of trying (I had to record at least an hour - usually the whole tape - with each setting to validate) I hadn't found it so was relieved when iuVCR/iuVCS did it first go.

Finally, going back to my question about archive format. Whilst at the moment I'm keeping the iuVCS AVIs, what is the best archive format. In 2013 lordsmurf preferred broadcast bitrate MPEG-2 for archiving. Is that still the case? MPEG-2 suits me fine as I can have DVR-MS files without recoding. But what bitrate and which MPEG-2 converter should I use? I already use VideoReDo TV Suite V5 a lot so am very familiar with it. Would this be suitable or do I need to use something else?

Again, many many thanks to you both for your suggestions. Now I've got some work to do...

John

[sanlyn]

If iuVCR gave you acceptable captures, that's what you should use. As I said, VirtualDub can have mysterious problems making nice with some system setups and non one seems to have the answer. AmrecTv is another lossless alternative.

Quote:

Originally Posted by jwhittin

Sorry for the delay - I've been offline for two days for various reasons.

Think nothing of it. I was late responding myself because I'm stuck with a klutzy Dell laptop until my home repair is finished and my PC's are set to go again.

Quote:

Originally Posted by jwhittin

no processing by me or anyone else after taking off the PAL VHS tape. So I'm a bit confused by the suggestion at least one is already de-interlaced. The wedding videos were captured in 1999 on "digital" by a semi-pro so I suspect it may have been a DV camera with the results copied onto VHS (I've got three copies). Surely he wouldn't have de-interlaced before saving it to VHS

Transfers from Dv to VHS were often done in the 90's and onward, mostly because DV wasn't designed as a distributable format, so VHS copies were made using various means. Just how that particular person worked the transfer is unknown, but it's definitely that the wedding video has been processed in a computer in various ways.

Well, you're correct, the outdoor wedding sample is interlaced. At first it didn't appar to be as the camera moves in slowly, but as people move and one gets deeper inbto the scene, the video shows as interlaced. In my haste I didn't go far enough into the sample in VirtualDub and came away thinking at first that it looked progressive. Blame my impatience for that one, because I didn't scroll far enough into the sample. But there has been processing of some kind and the indoor and outdoor shots are processed differently. For one thing, DV is almost always bottom field first (BFF) while VHS is almost always Top Field First (TFF). Both wedding samples are TFF. One is purposely blurred, the other is oversharpoened, and both are characterized by an absence of fine detail. The shot that "looks like" it would have been a DV original is the outdoor scene, which appears to have characteristic DV edge sharpening with a deficit of fine inner detail between edges. In any case, it's easily possible to reverse field priority in processing by deinterlacing and reinterlacing in a certain way. Changing it properly is harmless. Also keep in mind that DV is not always BFF.

You can easily check interlace or telecine and tracking effects by opening a file in VirtualDub and applying its built-in deinterlace filter. The photo below shows how the filter is set up.



- In the left-hand section nof the dialog, set "Deinterlacing mode" to "Interpolate using yadif algorithm".
- In the right-hand "Field Order" section, set "Double frame rate, top field first".

When you close this dialog, load the Indoor Wedding sample avi. Then use the navigation icons in the bottom left-hand corner of Virtualdub to view the video one frame at a time. We know that your door shot sample has 121 interlaced frames. But with the deinterlace filter mounted, you can see by the frame counter that the frame count is doubled to 242 frames after deinterlacing.

Now advance the video by one frame (one click of the ">" icon). The left-hand "Input" panel won't change, but in the right-hand "Output" panel the image moves forward by one field. Now click "> again and the Input and output panels both will move forward one field into the next frame. Click ">" again and the input panel remains the same but the output panel moves ahead one field. With each click of the forward or reverse button, the input panel remains on the same frame for two clicks but the output panel moves with every click of ">" or "<".

Next, load the Outdoor Wedding sequence. When click ">" or "<" the Input and Output remain the same for the first dozen frames or so, but going farther into the video you'll see by the people moving that it's interlaced video. I just didn't go far enough into the video to pick up on that. Meanwhile, using this frame-by-frame method you can also see how the lines between the church bricks wiggle and distort with each field, even though their basic positions stay the same with each field in the frame. When this tape is played at normal speed, that wiggling is obvious and annoying.

Using this frame by frame method, you can also see how uneven tracking affects objects in the Rio samples, especially the sample showing trees along the shoreline. You'll also be able to discern the luminance flicker and "blinking" of objects with each field in a frame. This would be a tracking issue. It's possible that the Rio tape has some damage or was mistracking in its maker's machine. You can try another tape to see if the geometry problems occur everywhere, but I don't see the same problems on the wedding tapes.

It appears that your 6-head player has a built-in line tbc, which is an essential requirement for tape capture. Likely your 4-head player won't have that feature but you can always try it.

Should you have any problems getting Avisynth and VirtualDub into working order, don't hesitate to ask. There are several standard filters in both apps that prove quite useful for a wide range of users. One popular Avisynth plugin is QTGMC. it's a deinterlacer, but it comes with several support filters that are useful and popular in their own right and are used by other plugins. VirtualDub also has some goodies that are in popular use.

Good luck with the project. There are always glitches during startup, so don't think that the rest of us haven't been there. This is very familiar territory.

Archiving:
Many will archive video as either original lossless versions or high-bitrate encodes. Normal DVD bitrates are from 5000 to 7500 mbps target bitrates. The absolute max bitrate for DVD is about 9000 kbps, which includes some leeway for the audio. Lordsmurf's reference to broadcast bitrates re in the range of 15,000 mbps for MPEG encoding. Some use high-bitrate h.264, which would be in the target range of 10,00 mbps.

Remember that lordsmurf is using a pricey MPEG encoder from MainConcept which is sold over here in the U.S. at $300 USD, and there are pricier versions. The free HCenc encoder can encode at BluRay MPEG bitrates, which are in the same bit range (but the HC encoder doesn't include audio). There are free encoders that use the free x264 engine for h.264 encodes at any bitrate you want, but anything 8000 mbps or higher with h.264 would be OK for archives. Keep in mind that MPEG and h.264 are lossy codecs, but with higher and higher bitrates you approach lossless quality. They aren't really designed for restoration work without quality loss, but they'll do in a pinch if you're careful.

[jwhittin]

Just to let you know that I'm working on it, experimenting with settings, and the initial tests look good and have taught me a lot about the filters. So many thanks again to you both.

It will take a while though. Even the "super fast" Wedding Indoor script is doing only 3-4fps on the PC I use for video (Phenom II X2 550 - definitely not a powerhouse but has been sufficient for all my limited video work to date). That's just over a day to process just one of the wedding videos...