Flickering colours on VHS tape?

[koberulz]

YRGBWaveform.jpg

This is a different clip, but the same principle applies: looking at the YRGB waveform in Color Tools, the blue channel is clearly clipping. In this case, using TweakColor to desaturate blue does work, as does ColorYUV working on the U channel.

However, using Histogram("levels") and Histogram("color2"), there doesn't seem to be any YUV clipping.

So when working on this, is it better to tackle it with TweakColor, or ColorYUV? If the latter, offset, or cont? Or is that a case-by-case thing?

Should I be aiming to drop blue below 255, or below 235? There are retrievable details above what shows up in that waveform, the clipping isn't occurring in-camera.

I'm still curious about the 16-235/0-255 setting in Color Tools. That screenshot was taken with it at 16-235. If I set it to 0-255, the narrow black bars on the left and right of the picture sitting at 16Y jump up to sit on the 31Y line, with a matching drop in the brights. The RGB channels are identical in both cases. Where should my white and black points be? It appears that in 0-255, I should be setting black at 16 and white at 235, but in 16-235 I should be setting black at 0 and white at 255.

EDIT: Ignore the junk on the right of the waveforms, I still had Histogram() active when I took the screenshot.

EDIT 2: The source for those waveforms. It's been through some processing, not an original, but nothing major (levels and QTGMC, basically).

You must be logged in to view this content; either login or register for the forum. The attached screen shots, before/after images, photos and graphics are created/posted for the benefit of site members. And you are invited to join our digital media community.

[sanlyn]

Quote:

Originally Posted by koberulz

This is a different clip, but the same principle applies: looking at the YRGB waveform in Color Tools, the blue channel is clearly clipping. In this case, using TweakColor to desaturate blue does work, as does ColorYUV working on the U channel.

Both work, yes. Reducing contrast in the U or V channel amounts to the same thing as reducing contrast in those channels. But remember that "cont" shrinks the histogram toward the middle from both ends, and the U channel also blends into yellow. ColorTweak or Tweak is more color specific. Reducing "gain" can also curb the brightest portion.

Quote:

Originally Posted by koberulz

So when working on this, is it better to tackle it with TweakColor, or ColorYUV? If the latter, offset, or cont? Or is that a case-by-case thing?

case by case, whatever works best.

Quote:

Originally Posted by koberulz

Should I be aiming to drop blue below 255, or below 235? There are retrievable details above what shows up in that waveform, the clipping isn't occurring in-camera.

y=16 becomes RGB=0 in RGB, y=235 becomes RGB 255 in RGB. That's the difference between the way luma and chroma are stored in video YUV and the way they display in RGB.

Quote:

Originally Posted by koberulz

I'm still curious about the 16-235/0-255 setting in Color Tools. That screenshot was taken with it at 16-235. If I set it to 0-255, the narrow black bars on the left and right of the picture sitting at 16Y jump up to sit on the 31Y line, with a matching drop in the brights. The RGB channels are identical in both cases. Where should my white and black points be? It appears that in 0-255, I should be setting black at 16 and white at 235, but in 16-235 I should be setting black at 0 and white at 255.

RGB is 0-255, regardless of those markers, which I usually ignore anyway. Visually you have to remember that RGB-0 displays as a really dark, dank entity, and RGB 255 is almost always too bright for something like a white shirt or jersey and "looks" overly bright. If you look at professionally produced digital video in RGB you'll see that a guy's black suit stops at about RGB 8 in its darkest shadows and most of the brights don't exceed RGB 245 not counting a few stray specular highlights. The numbers themselves don't always dictate the final visual result, except within legal video limits. The markers in ColorTools determines what level of pixels you want affected in the display of "hot" pixels that you want to see as black.

In encoding, the math for converting RGB to YUV performs the operation of contracting 0-255 to 16-235, if the program does it correctly.

The RGB waveform shows Blue as barely touching the top of the band. If you want to exercise very precise control on overall luminance or specific color(s), use the curves filter to form a notch cutoff at a certain point, shown here for Blue at the top right-hand corner of the diagonal line:



A similar correction is also available in most advanced color controls from ColorFinesse (AfterEffects), PremierePro, DaVinci, and others, as an auto preset of one kind or other. ColorFinesse calls it the NTSC safety slope, or something like that, when it's needed. It can be applied to single colors or to overall luminance. The extra anchor points (small circles) shown in the above image are used to keep values at their original points up to about RGB 226, above which the values gradually taper off.

[koberulz]

Quote:

Originally Posted by sanlyn

Reducing contrast in the U or V channel amounts to the same thing as reducing contrast in those channels.

Typo?

Quote:

But remember that "cont" shrinks the histogram toward the middle from both ends, and the U channel also blends into yellow. ColorTweak or Tweak is more color specific. Reducing "gain" can also curb the brightest portion.

Gain seems to just shift the whole thing across, and to unclip blue I have to add an awful yellow cast to the entire thing. Cont, set to unclip blue, results in U and V looking very similar, rather than the pre-adjustment histograms which are much wider in U than in V.

Quote:

RGB is 0-255, regardless of those markers, which I usually ignore anyway.

So why does the luma waveform change shape when I change which it's set to?

Quote:

The RGB waveform shows Blue as barely touching the top of the band.

There's a lot of detail recoverable via ColorYUV that isn't recoverable via gradation curves.

[sanlyn]

Quote:

Originally Posted by koberulz

Quote:

Typo?

Yep, sure is. Change that to:

Quote:

Reducing contrast in the U or V channel amounts to the same thing as reducing saturation in those channels.

Quote:

Originally Posted by koberulz

Gain seems to just shift the whole thing across, and to unclip blue I have to add an awful yellow cast to the entire thing. Cont, set to unclip blue, results in U and V looking very similar, rather than the pre-adjustment histograms which are much wider in U than in V.

Yes, that's what gain does, and that's the way it's explained in the Avisynth html. And cont shrinks from the ends toward the middle if negative, and from the middle outward if positive. In both cases the effect is stronger on the bright end.

Quote:

Originally Posted by koberulz

So why does the luma waveform change shape when I change which it's set to?

Did you notice that the small sub-panel in the ColorTools dialog is labeled "Source Attributes"?

[EDIT] Or to put it another way, the setting 0-255 turns off luminance expansion and 16-235 turns it on.

Quote:

Originally Posted by koberulz

There's a lot of detail recoverable via ColorYUV that isn't recoverable via gradation curves.

Unsafe clipping levels in YUV aren't recoverable in RGB, period. That's been covered several times. In YUV you can also use Levels, Y-Levels, Tweak, or SmoothLevels.

[koberulz]

Quote:

Originally Posted by sanlyn

Did you notice that the small sub-panel in the ColorTools dialog is labeled "Source Attributes"?

Yes, that's where the radio buttons are that I've been talking about.

Quote:

Or to put it another way, the setting 0-255 turns off luminance expansion and 16-235 turns it on.

So I should be setting the black point to 0 for 0-255, and 16 for 16-235? And the equivalent for the white point?

Quote:

Unsafe clipping levels in YUV aren't recoverable in RGB, period.

Right, so I'm not sure why you were suggesting gradation curves.

[sanlyn]

Set the radio buttons fro 16-235 to see luminance expansion at both ends.
Your own color corrections in RGB could result in unsafe values. You can use any number of filters to control that, gradation curves being one of them. Those filters are useful for keeping RGB values in tow, not for for recovery of unsafe YUV details.

[koberulz]

The fun part is when you're trying to color correct in Premiere and you realise you've screwed up what you tried to do in AviSynth and have to re-do a full day's worth of encoding. And then you leave it to run overnight and wake up to find the computer rebooted before it finished. *sigh*

Working on another tape, though, and the color correction tools are refusing to cooperate. Generally what I do is throw on an RGB Color Corrector set to display 'Tonal Range' to help pick out neutral brights, mids, and darks. Then use a garbage matte to isolate an area, turn on the RGB Parade, and (with the RGB Color Corrector turned off, obviously) adjust the relevant Three-Way Color Corrector wheel until red, green, and blue are equal.

For some reason on this tape everything is in the midrange according to the RGB parade, even though the RGB Color Corrector shows brights and darks. When I'm adjusting the brights and darks I can visually see the area I'm working on changing color, but the RGB Parade does. Not. Move. So I'm left having to guess at the brights and darks and the whole thing is jumping between really cool and really warm.

[themaster1]

try the plugin rgbadapt first (avisynth) then eventually tweak the colors manually. It's proven useful to me with awfully (de)colored sources

[koberulz]

What does that do that working with the three-way corrector in Premiere doesn't?

[sanlyn]

RGBAdjust is the RGB version of ColorYUV. It has functions just like those in ColorYUV but works on Red, Green, and Blue instead of Y, U, and V. It can't do any better than ColorYUV. But you're apparently talking about some specific color problems without posting samples, so guessing is all anyone can do.

[themaster1]

Quote:

Originally Posted by koberulz

What does that do that working with the three-way corrector in Premiere doesn't?

rgbadapt (not rgbadjust..different filter) does fix the video "automagically" so that will help you to have the video decently toned and if it's not enough work manually to fine tune.

plenty of examples on this page.
(note: check the last page of the thread to get the latest iteration of bernard's rgbadapt script)

[sanlyn]

Oh.
RGBadapt.
True, Master1, that's not the same as RAGBadjust.
My bad.

[koberulz]

So I FINALLY have a working ES10!

Pulled out this tape and tried it in my other two machines, and it seems like the Panasonic was partly responsible for the chroma clipping. It's ridiculously saturated regardless, but the sharp edges aren't there with the JVC or the Philips when looking at the Color2 Histogram in AviSynth.

[koberulz]

Through my JVC machine and the ES10.

[sanlyn]

The capture looks a little dim and lacking in dynamic range, but that could have been the capture settings. There is less chroma flicker but still some unnatural red blooming, likely on the tape rather in the electronics. The video needs a mild contrast and midtone boost.

The first script I ran was lordsmurf's contribution for fixing dropouts. It stabilized the jittery frames except for the frame shift just past the middle of the video. This was corrected later with RepaceFramesMC2.

The stabilization and major denoising was Script A, attached.

Script B tweaked saturation, bleeding, chroma shift, color, and dynamic range.

VirtualDub filters were used to curb some slight red and blue overshoot, to expand midtones, and to add some mild blue to yellowish skin tones. The vcf is attached. Filters used were ColorMill and gradation curves. The .vcf is also set to save the Avisynth results of Script B as Lagarith YV12 for encoding, which you can change if you wish.

Make script mods as desired.

The results are attached as an MP4.

[koberulz]

I increased brightness, decreased contrast, and massively decreased saturation for that capture. My previous capture had a larger brightness boost and no change to contrast or saturation.

[sanlyn]

Reducing capture saturation made that red jacket much easier to handle.

The first few frames have more noise than later ones, but that's common with temporal filters because with the first frames there are no previous frames to judge by. Sometimes it helps to precede the first frames with a few duped frames to force-prime the filter, so to speak.

[koberulz]

Quote:

Originally Posted by sanlyn

Reducing capture saturation made that red jacket much easier to handle.

That was the idea. I used GraphEdit to bring the footage straight from the capture device into an AviSynth script with a Histogram() and a Histogram("mode=color2"). The Panasonic, on which I ran my first capture, showed a very sharp corner in the bottom left of the vectorscope. The JVC and the Philips didn't. Both were well outside the circle, though, so I just dropped the saturation in the proc amp controls until the circle contained at least the vast majority of the color.