Flickering colours on VHS tape?

[sanlyn]

Quote:

Originally Posted by koberulz

Right, and the 'export, import' is a shift between two different passes.

What does "import,export" refer to? Why is it between passes?

If lordsmurf's script removes comets and dropouts, and is obviously faster than RemoveSpots and does some smoothing, why use RemoveSpots? BTW, I had trouble getting that script to work because of other plugins I didn't have at the time, so didn't get into it earlier. But tried it later after updating my other plugins. It has its good points, but so does RemoveSpots. It takes some experimentation to learn the pros and cons of different filters that do similar things.

Some filters work with SeparateFields, others (like FixChromasBleeding, Chromashift, etc.,) work best with full deinterlace. SeparateFields has one field displaced lower than the other, deinterlacing doesn't. Filters like RemoveSpots, RemoveDirtMC, mvDegrain2 work OK either way. Color correction doesn't care whether it's interlaced or not. Yes, deinterlacing does have its bad side (QTGMC can create artifacts like posterizing and over-cleaning if you're not careful), yadif doesn't denoise or do motion interpolation, a simple bob can soften the video, etc. ReplaceFramesMC should work on deinterlaced, but sometimes you might have to accept a slight blur on a frame or two if you use it on interlaced material, and if there's a lot of motion it can create bizarre effects.

You soon develop a knack for seeing what a filter does or doesn't do under different conditions. I usually save color correction for last because I don't care for going repeatedly between YUV and RGB back and forth.

For the attached samples I used two passes, all of it in Avisynth before going to a VirtualDub tweak. Part 1 is lordsmurf's NR and dropout cleaner. This cleaned dropouts and streaks and a little of the of the flicker. Results attached as studio1b Part 1.mp4.

Part 2 is for Avisynth levels and hue tweak (I left the histogram lines commented-out to show how I checked levels), QTGMC for denoising and shimmer and flicker reduction, fixes for chroma bleed and displacement, temporal smoothing for luma flicker, sharpening, adding some film-like grain to reduce over filtering effects. Then there's one line to repair the frame shift at frame 17 - 19, done while interlaced to avoid having to correct 6 shifted fields -- there's not much motion or combing, so this worked OK. Then RGB conversion for Virtualdub filters. I added ColorMill with a little midtone red and blue to keep the guy from looking too green. The results were saved as YV12 out of Virtualdub for encoding. Results attached as studio1b Part 2.mp4.

[koberulz]

Are you able to attach the VirtualDub settings file?

I probably wouldn't use RemoveSpots() if lordsmurf's script did enough. At present it's field-separated, working on even fields and odd fields distinctly, then an interleave and weave. I'm not sure LS's is faster, but it seems to have done a better job from the small sample I've been able to look at after running it first, followed by the same process I'd used initially.

Export/Import is exporting as an AVI in VDub, then bringing the result into a new AviSynth script.

I'm not familiar with SmoothTweak, ChromaMotion, DenoiseMC, GrainRestore, TemporalSoften, or grainfactory.

You usually use AddGrain(), I assume grainfactory is basically the same thing, any reason to choose one over the other? Given how soft the whole thing is to begin with, and how much denoising is going on, a softener seems an odd choice.

I think I asked this earlier and I think it got overlooked, but is an RGB conversion in the final AviSynth file necessary? Won't VDub just do that automatically if it needs to?

I've also found that using AddBorders in AviSynth doesn't really work, because they end up being affected by color correction. Not sure if I'm missing something, but it's easy enough to just add them as the last VDub filter.

I'm guessing the audio glitch in your sample file is ReplaceFramesMC screwing up? Is there a way to pull the audio out at the start of a script and then add it back at the end to avoid that? I've had to knock out 16 fields on a different clip and it's an absolute disaster audio-wise (video looks great, though).

[sanlyn]

Quote:

Originally Posted by koberulz

Are you able to attach the VirtualDub settings file?

All I did with ColorMill was go into the "RED-BLUE-GREEN" panel, select "Mid" for midtones, then raised Red and Blue each by 5 points. VCF attached.

Quote:

Originally Posted by koberulz

I'm guessing the audio glitch in your sample file is ReplaceFramesMC screwing up? Is there a way to pull the audio out at the start of a script and then add it back at the end to avoid that? I've had to knock out 16 fields on a different clip and it's an absolute disaster audio-wise (video looks great, though).

OOps! Drat, I almost always forget about that myself. Serves me right for eing too lazy or sleepy to turn my speakers on.


There are two ways to fix it. You can save the audio in a script by using these lines:

AviSource("path to video"\video.avi")
aud=last   #<- saves a copy of the source video, including audio

last       #<- return focus to the AVI Source.
...some processing...
...more processing...
...down to end of script...
vid=last
AudioDub(vid,aud)
return last

AudioDub() is an Avisynth built-in function. Another way of doing it is to load the AVI into Virtual with no script or processing. Use "File..." _Save WAV...", give the .wav file a name and location, and save. Omit the "aud=last" at the beginning of the script. Instead, load the .wav file into the script later at the end:

vid=last
aud=WavSource("path to file\audio file.wav")
AudioDub(vid,aud)
return last

Be careful that the .wav file you save doesn't belong to an AVI that is going to be trimmed later during the script. If the script changes the number of frames in the AVI, audio won't match. One way around this with a Trim is to use an AVI that's already trimmed, or do it this way:

AviSource("path to video"\video.avi")
Trim(8902, 12894)
aud=last   #<- saves a copy of the source video, including audio

last       #<- return focus to the AVI Source.
...some processing...
...more processing...
...down to end of script...
vid=last
AudioDub(vid,aud)
return last

Quote:

Originally Posted by koberulz

I'm not familiar with SmoothTweak

It's a function in the SmoothAdjust plugin. If you look at its Readme doc it's explained in great detail.

Quote:

Originally Posted by koberulz

ChromaMotion, DenoiseMC, GrainRestore

They're options in QTGMC. They're explained in the html that comes with the QTGMC plugin. Attached.

Quote:

Originally Posted by koberulz

TemporalSoften

It's an Avisynth bullt-in. It's more a luma smoother like TTempSmooth than it is a softener, but not as aggressive. http://avisynth.nl/index.php/Soften

Quote:

Originally Posted by koberulz

You usually use AddGrain(), I assume grainfactory is basically the same thing, any reason to choose one over the other? Given how soft the whole thing is to begin with, and how much denoising is going on, a softener seems an odd choice.

They don't smooth noise, they add dithered noise. I use AddGrainC, not AddGrain. GrainFactory has slightly coarser grain, is more movie-like. Both are used to mask hard edges in gradients and to avoid an over-smoothed, over-filtered appearance. They're similar. Use the one you think looks best for the job.

Quote:

Originally Posted by koberulz

I think I asked this earlier and I think it got overlooked, but is an RGB conversion in the final AviSynth file necessary? Won't VDub just do that automatically if it needs to?

On input you get cleaner RGB conversion if you do it in Avisynth. On Virtualdub output, set color depth to what you want, then set the compressor to use that colorspace. That output conversion is usually final, so little or no harm done.

If you're not applying VirtualDub filters and you want to avoid RGB and keep the source colorspace and compression, save using "fast recompress". That mode won't apply any Virtualdub filters, though, even if they're loaded.

Quote:

Originally Posted by koberulz

I've also found that using AddBorders in AviSynth doesn't really work, because they end up being affected by color correction. Not sure if I'm missing something, but it's easy enough to just add them as the last VDub filter.

I do it in Avisynth to center the image in the frame. If you need to mask it over because of color work later, go ahead. Unless you've had to change RGB-zeo to a lighter color, the added border is RGB 0 0 0 (dead black = no color).

I corrected the mp4's audio for that glitch (attached). I used the "aud=last" method, but I also do it the "Save wav...." way half the time.

[koberulz]

Quote:

They don't smooth noise, they add dithered noise. I use AddGrainC, not AddGrain. GrainFactory has slightly coarser grain, is more movie-like. Both are used to mask hard edges in gradients and to avoid an over-smoothed, over-filtered appearance. They're similar. Use the one you think looks best for the job.

Sorry, I just sort of ran through everything I didn't know from that script all mashed together (and went from memory on AddGrain). The comment about it being soft and such was in reference to TemporalSoften(), not grainfactory().

The aud=last method seems easiest; I'll probably keep the RemoveSpots script (which is now badly named, it's just frame replacements now) as the last step in every process anyway so it's already there.

[sanlyn]

RemoveSpots and its variations do more than just remove spots, they are like pumped-up versions of RemoveGrain and RemoveDirt. The "MC" versions don't soften as much as non-MC. Watch out for them in animation and other fine-detail videos, as they can often remove tiny thingies that are really details and can break lines in animation. The lordsmurf big script is a more thorough denoiser for bad shimmer in frames. I used it for a partial fix in this thread: http://www.digitalfaq.com/forum/vide...avy-lines.html.

[koberulz]

I meant I had a RemoveSpots script, which ran...well, RemoveSpots, obviously. But now that I've run lordsmurf's script, I'm not running RemoveSpots, that script is just running ReplaceFramesMC to clean up what lordsmurf couldn't. I've just gone through and removed six or seven instances that are no longer required. The 16-field issue is still there but it's gone from being absolutely awful to something you could almost get away with not touching.

Manually watching the whole thing through to note field numbers, though...ugh.

[lordsmurf]

The script I gave assumes interlaced source, since it's been written for VHS. But I've run it on progressive source, and allowed it to remain as interlace. And at least once, I've run it on progressive, then QTGMC the output. Some purists may balk at this, but after much testing, that was the best result for the specific project. The mere act of deinterlacing can have beneficial effects, though rare.

[koberulz]

Quote:

Originally Posted by sanlyn

RemoveSpots and its variations do more than just remove spots, they are like pumped-up versions of RemoveGrain and RemoveDirt. The "MC" versions don't soften as much as non-MC. Watch out for them in animation and other fine-detail videos, as they can often remove tiny thingies that are really details and can break lines in animation.

On this subject, lordsmurf's script has a habit of removing the entire basketball if it's thrown fast enough. Which is pretty neat, in its own way. Does turn it into a bit of a balancing act though.

Can you explain the use of TemporalSoften() in your last attached script? I've read the wiki page and I kind of get the technical aspect, but not really how it applies to this particular clip.

How does using SmoothTweak differ from levels adjustments and ColorYUV?

[sanlyn]

Quote:

Originally Posted by koberulz

On this subject, lordsmurf's script has a habit of removing the entire basketball if it's thrown fast enough.

That's a brute force anti-dropout anti-comet filter not designed for routine use. I wouldn't use it on video with lots of fast motion, for the reason you described..

Quote:

Originally Posted by koberulz

Can you explain the use of TemporalSoften() in your last attached script? I've read the wiki page and I kind of get the technical aspect, but not really how it applies to this particular clip.

It's for the red choma and chroma flicker leftovers in that red shirt. There's also some horizontal tape noise visible in brighter areas and the face. It also helped smoothed some hard contours in the face.

Quote:

Originally Posted by koberulz

How does using SmoothTweak differ from levels adjustments and ColorYUV?

Main avantage is smoother dithering in all of SmoothAdjust's functions. Try adjusting contrast and saturation using ColorYUV and watch gaps and rough spots show up in the histogram. To do what SmoothTweeaks' Contrast setting does, you can't quite get there with ColorYUV -- its "gain" parameter is the nearest operator but it affects darks whereas SmoothTweak's contrast doesn't. ColorYUV's contrast settings tend to flatten midtones and has no dithering to smooth the stretching or contracting effects. You can always try getting the same results using any two of the filters mentioned, or just one, if you wish. Otherwise you're comparing three different filters designed for different sets of operations.

Remove TemporalSoften, SmoothTweak, SmoothLevels, and ColorYuv. Then start from scratch with whatever you want. Avs scripts aren't scripture. There are many filters that can accomplish the same things.

[koberulz]

Quote:

Originally Posted by sanlyn

That's a brute force anti-dropout anti-comet filter not designed for routine use. I wouldn't use it on video with lots of fast motion, for the reason you described..

Given the amount of dropouts, it's preferable to trying to go through with ReplaceFrames. I ran one with, and one without, and I'm using Premiere to switch between them. I'm 35 minutes into the 1:07 tape, and the ball has disappeared for two frames. Meanwhile, all the other issues have been almost entirely eliminated.

Quote:

To do what SmoothTweeaks' Contrast setting does, you can't quite get there with ColorYUV -- its "gain" parameter is the nearest operator but it affects darks whereas SmoothTweak's contrast doesn't. ColorYUV's contrast settings tend to flatten midtones

Could you explain this a little more? The documentation for SmoothAdjust merely says contrast alters contrast, which...well, obviously.

Quote:

Avs scripts aren't scripture. There are many filters that can accomplish the same things.

No kidding, I think this is the fourth different script someone's done for this clip.

I'm asking questions because I want to learn. I've got a lot of tapes to do here, I can't just put them all up here and say 'write a script for me'. Well, I could, but I don't think it would go down very well.

Unfortunately that's led to a bit of meandering, as I'll see something used in this discussion and realise it would be really good for a different tape (setting colors/levels in YUV for that blown-out color-cast one, dropout fixes for the other one I've been working on).

I've just realised after all the hassle of trying to get MCTemporalDenoise to work, this most recent script doesn't even use it!

[sanlyn]

Quote:

Originally Posted by koberulz

Given the amount of dropouts, it's preferable to trying to go through with ReplaceFrames. I ran one with, and one without, and I'm using Premiere to switch between them. I'm 35 minutes into the 1:07 tape, and the ball has disappeared for two frames. Meanwhile, all the other issues have been almost entirely eliminated.

You're referring to filters that target different problems but have different side effects. ReplaceFrames interpolates new motion from preceding and following frames, MCTD in chroma-only mode smooths chroma noise between frames but not motion. MCTD smoothed chroma flicker but didn't affect the ball. Replace frames creates new motion for a frame but not without creating either mild or severe physical artifacts, sometimes stretching or distorting objects, sometimes removing them. The anti-dropout script could well have decided that the ball was part of a dropout and smoothed its contours into other objects. Most filters have more precise targets than others, which is why there are so many filters for so many different problems.There is no one-size-fits-all.

Quote:

Originally Posted by koberulz

Could you explain this a little more? The documentation for SmoothAdjust merely says contrast alters contrast, which...well, obviously.

ColorYUV's contrast functions tend to stretch the histogram at both ends, making darks darker and brights brighter but thinning out the histogram band. Negative contrast values with ColorYUV tend to contract the histogram toward the middle, lightening darks and darkening brights. ColorYUV's gain function applies a multiplier to the band; positive values multiply darks by a small amount, but multiply brights by increasing amounts. Positive gain tends to stretch the histogram forward from the darks outward, negative gain pushes the band leftward from the brights toward the darks. SmoothTweak's contrast stretches the histogram from the middle outward, while negative values contract the histogram from the brights toward the middle. So which one you use depends on the effect you want. With ColorYUV you can work on the contrast of luma and the chroma bands separately; SmoothTweak contrast works with luma only. SmoothTweak has no separate gain function. You can observe the difference in effects by watching a histogram as you change filters and values.

Quote:

Originally Posted by koberulz

I've just realised after all the hassle of trying to get MCTemporalDenoise to work, this most recent script doesn't even use it!

You're forgetting why MCTD was used in the first place. Lordsmurf's script was designed for dropouts, not for chroma flicker. The script did attack that flicker mildly (than kjs to its use of RemoveGrain using its diffrerent mode settings) but other filters were still required for that flicker and shimmer cleanup and for other noise problems. MCTD didn't clean dropouts but it didn't remove the ball either, and neither did RemoveSpotsMC. Neither MCTD nor RemoveSpots created distortion on movement, but ReplaceFrames did and isn't usually recommended anyway where lots of motion is involved. Each of the filters had different effects on different issues. If you have to apply two or three different filters for two or three different problems, so be it. There's no one filter for everything.

[koberulz]

Sorry, the ball-removal was a different game that sprung immediately to mind when I saw what the lordsmurf script could do. This thread has become a bit muddled, which is why I created a second thread rather than bringing that issue in here.

Quote:

Originally Posted by sanlyn

You're referring to filters that target different problems but have different side effects.

I didn't mention, or even really think about, MCTD until the last line of that post. My point was that dropout removal via lordsmurf's script, cutting between a version that had undergone that process and a version that hadn't, was easier than dropout removal via ReplaceFrames due to the high number of dropouts and therefore running the whole thing through that script was worthwhile. The chroma noise, as you note, is a separate issue entirely.

Quote:

The anti-dropout script could well have decided that the ball was part of a dropout and smoothed its contours into other objects.

I have no doubt that it did, and in fact it wasn't entirely unexpected. It's actually pretty impressive, given that RemoveSpots sometimes seems to do almost nothing.

Quote:

ColorYUV's contrast functions tend to stretch the histogram at both ends, making darks darker and brights brighter but thinning out the histogram band. Negative contrast values with ColorYUV tend to contract the histogram toward the middle, lightening darks and darkening brights...SmoothTweak's contrast stretches the histogram from the middle outward, while negative values contract the histogram from the brights toward the middle.

So ColorYUV contrast leaves 50 alone and moves everything else away from it, SmoothTweak contrast leaves 50 alone and moves only higher values away from it, and ColorYUV gain leaves 0 alone and moves higher values away from it, with an increasing effect the higher it goes?

Where 0 and 50 are percentages of full white, not Y values.

Quote:

You're forgetting why MCTD was used in the first place. Lordsmurf's script was designed for dropouts, not for chroma flicker.

Yes, I know. However, I was figuring you also attacked that rather than merely the dropouts, which given that response you may not have.

[sanlyn]

Quote:

Originally Posted by koberulz

So ColorYUV contrast leaves 50 alone and moves everything else away from it, SmoothTweak contrast leaves 50 alone and moves only higher values away from it, and ColorYUV gain leaves 0 alone and moves higher values away from it, with an increasing effect the higher it goes?

ColorYUV contrast stretches the histogram in both directions if positive, contract it toward the middle if negative. Gain multiplies all values from 0 upward if positive, contracts toward 0 if negative. SmoothTweak works from them middle and brighter of positive, and from brights toward the middle if negative -- dithers pretty smoothly which ColorYUV doesn't do at all.

Color YUV contrast:
    positive
    <------<|>------->

    negative
    >------>|<-------<


ColorYUV gain: (works like gain in TV calibration):
    positive
    >|>----- - - - - > 
    
    negative
    <|<----- - - - - <


Smooth Tweak contrast:
   positive
    |    |----------->

   negative   
    |    |<-----------

More or less.

[lordsmurf]

I think this thread is getting to be one of the longest non-sticky threads on this forum.

And it's chocked full of interesting info on each page. Nice.

[koberulz]

I've ordered a third ES10. Fingers crossed!

How does one work out saturation? I'm satisfied with the studio clip, so I'm working on the actual game. Obviously yellow and red are dominant, but given what's on screen 99% of the time, they should be. So I'm not sure how much I need to drop them. I can't seem to get red to stop clipping on ColorTools' RGB waveform monitor.

I'm also having issues with using the YC waveform monitor in Premiere. Apparently you're supposed to have the chroma cover roughly the same range as the luma, but I always end up with either a near-greyscale image (if I adjust master saturation), or clear delineations between the mids and highlights (if I adjust highlight saturation). Otherwise, chroma is miles above luma.

[sanlyn]

If you clean up oversaturation in Avisynth/YUV (16-235), they won't clip in RGB. I use Histogram(mode="Color2")

I can't answer for Premiere Pro's controls, I don't use it.

[koberulz]

I tried setting red saturation to zero with TweakColor(), and red was still clipping. Luma was fine, as were green and blue, so it wasn't that.

But beyond that, red and yellow seem overly strong. Is it just an 'adjust by eye' thing?

[sanlyn]

All I could do here is post the same techniques, images, and histogram captures and scripts that were already posted. I'm afraid that I can't accept "setting red saturation to zero with TweakColor(), and red was still clipping" as telling me that you undestand what has been discussed so far, and I don't know how to explain it to you without repeating vast portions of this thread. Reducing saturation doesn't reduce brightness, it reduces chromatic intensity. If you take an overly bright color and make it gray (zero chroma saturation), it's still overly bright. I have no idea what you're doing.

[koberulz]

Quote:

Originally Posted by sanlyn

If you clean up oversaturation in Avisynth/YUV (16-235), they won't clip in RGB.

Not sure how adjusting saturation is at odds with this.

[sanlyn]

Then you're not describing the problem correctly. If you have adjusted your chroma levels and luma levels in YUV so that they don't exceed 16-235, those values won't exceed 0-255 when expanded in RGB. But some of the stuff in your tape has clipping in-camera. You can't recover data that was destroyed in-camera or in subsequent processing.

Look back at the histograms posted in post #99 (http://www.digitalfaq.com/forum/vide...html#post46045):



The YUV Levels chart on the left shows an image with luma levels that exceed y=235 at the bright end. The values that exceed y=235 will be greater than RGB=255 and will be clipped when expanded in RGB. The histogram on the right measures chroma intensity or saturation, not luma brightness levels. The right-hand "Colors2" shows an image with under-saturated colors, no reds, mostly greens and cyans, which is what the dull-looking washed-out images of that video look like. They illustrate that you can have low saturation with clipped luma, and vice-versa. YUV stores luma and chroma separately, and they can be processed separately.