Hi8 artifacts (grain, color balance, noise at bottom) Plugins tailored for?

[Colmino]

Let me flesh out the thread subject a bit. I've got a bunch of Hi8 (and even some regular 8mm) tapes which were archived as DV via a Digital8 camcorder. They look about as bad as one would expect. The final version of the video project I'm working on is going to feature some of the footage from these Hi8 videos as part of a hi-def presentation (probably 1080i60). The poor quality of the video will be made very apparent.

Now then. I know there must be plugins and filters out there which were created to deal with typical analog tape noise, because everyone has old tapes they'd like to take a stab at improving. But there's plenty I don't know.

1) The pie-in-the-sky wish. Is there a plugin or package whose chief or sole design purpose is to analyze once-analog (now digital) video (e.g. DV 720x480) and spit out a cleaned-up rendition? Preferably for Premiere Pro, although After Effects is a dubiously acceptable alternative.

2) Barring an all-in-one, what are some of the better plugins out there for cleaning up analog artifacts? Doesn't have to be freeware. Strong preference is for something which works from within Premiere Pro or After Effects. If I have to dig out VDub, that will just add a big step, much complication, and probably a huge amount of raw video storage.

Now, what do I mean by "analog artifacts"? There are a few different and very familiar problems which I am seeing with my archived video, and indeed recall seeing when I used to watch things on VHS. I don't know the official terminology for any of them, so I'll have to make do with descriptions.

a) Noise. It's everywhere. It looks, really, a great deal like film noise.
b) Bleeding. Wherever there are sharp luma contrasts (such as, for example, embedded timecode), the color from surrounding imagery tends to bleed into the whiter areas.
c) Horizontal inaccuracy. Basically, from top to bottom, ostensibly vertical lines look more like heavily wavy lines, though you have to look close to notice.
d) Color blandness. Compared to any good DV camcorder, the color I'm seeing from these tapes is very bland. I could *manually* adjust it and achieve undeniable improvements that way, but this would be very time-consuming, especially in the beginning weeks while I was still figuring out what adjustments were commonly needed. This one I am overwhelmingly confident there has to be a plugin to handle such adjustments hands-free.
e) The ever-present skewed video at the very bottom. I don't think I've ever seen any analog tape, transfer or otherwise, that hasn't exhibited this. The bottom three or four lines of video are heavily skewed to one side, making them contextually indecipherable, although clearly still part of what was originally recorded. It seems to me that a smart plugin (or a patient person) could manually shift those lines back to their correct position and rescue at least some of that otherwise unusable video. Would love to know what this artifact is actually called.
f) Occasional moments of heavy noise in the bottom 10% or so. Bad tapes cause this. Heavy snow at the bottom which comes and goes, occasionally accompanied by drops in audio. I see this one frequently enough that I don't think it's unreasonable to anticipate a plugin that has a good solution. (Not the audio problem, obviously.)

I've eyeballed NeatVideo. It seems like a good, if nonspecific, solution for the general noise problem (#a) but I haven't seen anything which more particularly focuses on analog tape artifacts. AlgoSuite, a decidedly not-cheap set of tools, seems perfect for digital artifacts and deinterlacing, but that's not really what I'm after (though I would love to compare it with NeatVideo to see which is better at fixing the DV noise).

Anyway, guess I'll leave it at that. Thanks for any advice.

[CogoSWSDS]

Flaxen's fxVHS filter does a good job of cleaning up noisy video footage, although it is for VirtualDub and is very slow. A significantly faster filter (but not quite as good at cleaning up the noise and a bit more difficult to use, IMHO) for cleaning up noisy video footage is Convolution3D for AVISynth.

For Convolution3D, you have to separate the fields into two different clips, apply the filter to the clips separately, then recombine the clips back into an interlaced video. The script for doing this with AVISynth is easier than you might think.

LoadPlugin("D:\Path\to\Convolution3D.dll")

function SetParity(clip c, bool parity) {
    return parity ? c.AssumeTFF():c.AssumeBFF()
}

function c3d(clip c) {
    oldParity = c.getParity();
    c=c.AssumeTFF().SeparateFields().AssumeFrameBased();
    top=c.SelectEven().Convolution3D(0,6,10,6,8,2.8,0);
    bottom=c.SelectOdd().Convolution3D(0,6,10,6,8,2.8,0);
    c = interleave(top,bottom).AssumeFieldBased().AssumeTFF().Weave();
    return c.SetParity(oldParity);
}

The first line of the script loads the Convolution3D filter. The first function allows you to set whether the clip is Top Field First or Bottom Field First depending on the value of "parity".

Now for the meat of function c3d:
First line gets the parity of the clip (true if TFF, false if BFF).
Second line separates the fields with top fields first. If you're in NTSC land and your clip was 720x480 at 29.97fps, you now have a clip of 720x240 at 59.94fps.
Third and fourth lines separate the fields into two separate clips, "top" which contains all the even (top) fields, and "bottom" which contains all the odd (bottom) fields, and applies Convolution3D to both clips separately.
Fifth line weaves the two clips back together and re-interlaces them, top field first.
Last line sets the clip to be either TFF or BFF depending on the characteristics of the clip passed to it.

For an excellent guide on installing and using Convolution3D, check out this thread by our own FulciLives:
https://forum.videohelp.com/topic225951.html

Convolution3D can be downloaded from http://www.avisynth.org/warpenterprises/ while Flaxen's fxVHS filter for VirtualDub can be downloaded from www.neuron2.net. It is only fair that I warn you that while Flaxen's fxVHS filter does an excellent job of cleaning up noisy video, it is very slow. On my 2GHz single-core machine it takes 3 hours to clean up 1/2 hour of video. Convolution3D would only spend about an hour doing the same thing.

The Convolution3D will fix up the noise but leave the other issues. FxVHS will take care of the noise and the chroma (luma?) problems, which is a/b on your list. If you didn't already have the video converted to digital format, I'd say a TBC would do well to take care of issue c. You may be stuck with the waviness now that it's digital. I've seen amazing results from the VirtualDub filter "ColorMill" that would take care of the color blandness. As for the skewed video at the bottom, I'd overlay that with black (I'm assuming that's head switching noise). Overscan of the TV set would hide this from your view anyway, just as it did when you played it back from the original Hi8/Video8.

I don't think there's any way to handle issue f. Once the data's gone, it's gone.

I've heard good things about NeatVideo, but never tried it, so I can't give you an opinion on that one.

Hope this helps.

CogoSWSDS

[jagabo]

c) Horizontal inaccuracy. Basically, from top to bottom, ostensibly vertical lines look more like heavily wavy lines, though you have to look close to notice.

You'll need a time base corrector between the Hi8 player and the capture device to fix this.

[guns1inger]

1) No. There isn't a single plugin that will do all that you want. There are a wide range of plugins available for avisynth and virtualdub that cover many of your issues, however you will have to get your hands dirty. Neat video is quite good at what it does, albeit very slow to run, but for the majors, that is about it.

2. See answer to question 1)

a) NEAT is a good solution. With any noise reduction you need to be careful that you don't remove too much noise and leave plastic people in it's place. Some noise is natural and good.

b) and c) can really only be properly solved by using high-end transfer equipment. You will never get rid of them transferring through a camera the way you are, and once they it digital, it is too late. Some luma issues can be corrected with filters, but not as well as you will probably want. Also, some of the bleed is inherent in the recording and cannot be removed.

d) Colour correction and adjustment tools in Premiere or AE are as good as most

e) a TBC may clear some of this up.

f) sounds like tracking or tape damage. No filter to full repair.

A couple of basic issues need to be addressed before you progress too much further.

1. The best place to fix the bulk of your issues is before the digital encoding takes place. This means getting better transfer equipment, a time base corrector, and proc amp. Using a digital8 camera means that all the damage is done in the camera, and once it is digitised you severely limit
how much repairing can be done.

2. Video restoration is time consuming. Every video is different, and needs to be treated based on it's specific needs. This is one of the reason there isn't a simple, automated, one button process.

If you want to do this yourself, you have to be prepared to spend the time and money required to get the right equipment and learn all the tools, and then do the work, or be prepared to lower your expectations regarding the outcomes. Or, you can get a professional restoration house to do the work for you. This will be very expensive, but this may not be the main issue for you. When I say Professional restoration House I do not mean the dinky little outfits that advertise on the web with a garish website knocked up by their grade 6 nephew in frontpage, and covered in testimonials from happy customers. I mean a truely profession company that studios would use.

Restoration can be incredibly frustrating, time-consuming and difficult. You have to be prepared to put in the hard yards to do it properly, or even half decently. It is also incredibly rewarding when you see what you have achieved. But there is no simple way to get there.

[Soopafresh]

Neatvideo will probably be the solution closest to your requirements. It's very good - for the $ it's excellent, but just remember that nothing with make bad analog stuff look great in HD. It's as important to lower your expectations as it is to find the correct noise reduction plugin.

[2Bdecided]

You can reduce some colour bleeding in software:
http://forum.doom9.org/showthread.php?t=136373

Also, I think better equipment doesn't help that much with this issue given the pathetic chroma bandwidth on consumer video (all types!). That smeary noisy mess (have you ever tried watching just the chroma? yuk!) is all there is on the tape!


You do need a TBC to fix problem (c) - if you've dumped your original tapes, you'll have to live with what you have now because it's too late.

Most people blank the bottom few lines to fix (e). A TBC can help, but the results are still not usable most of the time. Video inpainting may help, but is still to slow and experimental IMO.

btw, I hope you transferred Hi8 via the S-video connector. Using composite throws a lot away.

Finally, whatever you do, it'll look terrible at 1080i60. If anything, all the noise you're trying to remove might help to hide the lack of detail, and hence be beneficial! Or you might choose to add a little noise back in at 1080i to fake hi-res detail. Strange but true.

Cheers,
David.

[fredfillis]

I've spent a lot of time capturing Hi8 and printing to DV tape. One thing I've noted is that the video plays worse on a monitor than it does on a TV. In my case, SD. But, I've saved myself a lot of work by not correcting my footage to look good on my PC. It might not be relevant in your case, but, have you looked at your DV footage on something else besides a PC?

[deuce8pro]

Nevermind a Hi8 player.
Get a Sony Digital8 cam with tbc/dnr that can play 8mm/hi8 tapes.
Get a firewire cable
and the JVC DRM100 DVD Recorder.
Encode in XP on this machine.
You'll have much better results.

Also if the tapes are old. Fast forward and rewind them twice before putting them in the d8 cam. Use your old camera. The tapes need
to be tight to play correctly.

Let me also add you will unlikely get the color bleeding problem if you use this method and the drm100 will help to clean up some of the noise on the tape.

[zoobie]

not sure it's realistic to go out and buy another cam and another DVD recorder
but running the tapes back and forth is good...it keeps them from binding

I, like others here, have noticed that the hi8 format actually needs it's noise to display properly...just how much is up to you

[zcream]

Algosuite has actually been discontinued - but if someone has a license I would love to buy it..

Flaxen's fxVHS filter does a good job of cleaning up noisy video footage, although it is for VirtualDub and is very slow. A significantly faster filter (but not quite as good at cleaning up the noise and a bit more difficult to use, IMHO) for cleaning up noisy video footage is Convolution3D for AVISynth.

For Convolution3D, you have to separate the fields into two different clips, apply the filter to the clips separately, then recombine the clips back into an interlaced video. The script for doing this with AVISynth is easier than you might think.

LoadPlugin("D:\Path\to\Convolution3D.dll")

function SetParity(clip c, bool parity) {
    return parity ? c.AssumeTFF():c.AssumeBFF()
}

function c3d(clip c) {
    oldParity = c.getParity();
    c=c.AssumeTFF().SeparateFields().AssumeFrameBased();
    top=c.SelectEven().Convolution3D(0,6,10,6,8,2.8,0);
    bottom=c.SelectOdd().Convolution3D(0,6,10,6,8,2.8,0);
    c = interleave(top,bottom).AssumeFieldBased().AssumeTFF().Weave();
    return c.SetParity(oldParity);
}

The first line of the script loads the Convolution3D filter. The first function allows you to set whether the clip is Top Field First or Bottom Field First depending on the value of "parity".

Now for the meat of function c3d:
First line gets the parity of the clip (true if TFF, false if BFF).
Second line separates the fields with top fields first. If you're in NTSC land and your clip was 720x480 at 29.97fps, you now have a clip of 720x240 at 59.94fps.
Third and fourth lines separate the fields into two separate clips, "top" which contains all the even (top) fields, and "bottom" which contains all the odd (bottom) fields, and applies Convolution3D to both clips separately.
Fifth line weaves the two clips back together and re-interlaces them, top field first.
Last line sets the clip to be either TFF or BFF depending on the characteristics of the clip passed to it.

For an excellent guide on installing and using Convolution3D, check out this thread by our own FulciLives:
https://forum.videohelp.com/topic225951.html

Convolution3D can be downloaded from http://www.avisynth.org/warpenterprises/ while Flaxen's fxVHS filter for VirtualDub can be downloaded from www.neuron2.net. It is only fair that I warn you that while Flaxen's fxVHS filter does an excellent job of cleaning up noisy video, it is very slow. On my 2GHz single-core machine it takes 3 hours to clean up 1/2 hour of video. Convolution3D would only spend about an hour doing the same thing.

The Convolution3D will fix up the noise but leave the other issues. FxVHS will take care of the noise and the chroma (luma?) problems, which is a/b on your list. If you didn't already have the video converted to digital format, I'd say a TBC would do well to take care of issue c. You may be stuck with the waviness now that it's digital. I've seen amazing results from the VirtualDub filter "ColorMill" that would take care of the color blandness. As for the skewed video at the bottom, I'd overlay that with black (I'm assuming that's head switching noise). Overscan of the TV set would hide this from your view anyway, just as it did when you played it back from the original Hi8/Video8.

I don't think there's any way to handle issue f. Once the data's gone, it's gone.

I've heard good things about NeatVideo, but never tried it, so I can't give you an opinion on that one.

Hope this helps.

CogoSWSDS

[sphinx99]

I digitized many terabytes of Video8 and Hi8 video a couple of years ago and was then confronted with the same questions you're asking today. My decision then, which remains unchanged, is that PC-based digital restoration continues to evolve, and the tools I'll have in hand 5 or 10 years from now will handily outshine the tools I have today.

So I've kept all my sources exactly as they are in their unrestored huffyuv/lagarith/dv-avi glory. Every now and then, I need to convert some to DVD for one reason or another, and I'll use whatever is the current restoration process fad of the month (largely driven by doom9) at the time to do it, but I leave the source untouched.

What I've noticed about 18-24 months in is that the tremendous increase in compute power lets me use more sophisticated denoisers and other filters than I had available to me 2 years ago. What was 1fps in '06 is 8fps or 10fps in '08 thanks to quad-core and SSEn and blah blah blah.

I know I'm not giving you a specific answer but rather some advice: if you don't have a pressing need to finalize your video, and if you have the storage capacity to preserve and maintain the original captured sources, then I'd suggest that you let sleeping dogs lie, and check back on this website or doom9 in 2010 so catch up on how things have changed.

[2Bdecided]

I'm keeping it all as untouched DV-AVI for now. I don't have the space for lossless.

However, while I agree that processing power increases every year, and clever new techniques are born at doom9 every day, there will come a time when fewer and fewer people are processing analogue video. At this time, new tools will stop being produced, and I bet the old ones won't run nicely on new hardware.

So don't leave it for 20 years!

Cheers,
David.