TCB-1000 blue screen noise

[SkyBlues2021]

Thanks Sharc. Sorry I didn't see your added improved clip. It looks like you've removed a mist from the video. Very impressive. What filters did you use to get that improvement?

[jagabo]

The main problem with your "mist" is a black level that's too high.

[Sharc]

It looks like you've removed a mist from the video. Very impressive. What filters did you use to get that improvement?

Yes, as jagabo wrote, the milky look of your capture is due to a too high black level. All your clips seem to suffer from this, probably because you capture over-cautiously, ensuring that everything is within the 16....235 luma range, i.e. including under- and overshooting artefacts and noise.
Anyway, it can be easily fixed in post processing:

Avisynth:

SmoothLevels(input_low=36, gamma=1.0, input_high=215, output_low=16, output_high=235, HQ=true, TVrange=True, Limiter=0) #decrease the dark levels and increase the brights somewhat
SMDegrain(interlaced=true) #for denoising (or use some other interlace-aware denoiser of your choice)

VirtualDub:
Video->Filters-> Levels .... pick the levels filter. Settings see attachment.
You may then want to denoise it as well using one of the VirtualDub's denoisers

(You can also use filters for brightness/contrast/saturation and achieve the same/similar result as with the levels filters).

[jagabo]

Histograms often aren't very good for setting levels. The problem is that you don't know what parts of the picture correspond to the levels:


[Attachment 66872 - Click to enlarge]

You can see that big peak near the left of the histogram. But if you crop away the black borders before getting the histogram:


[Attachment 66873 - Click to enlarge]

You can now see that it was those black borders that caused that peak. There's nothing very dark in the part of the picture you care about.

A Waveform monitor is better for this type of analysis:


[Attachment 66874 - Click to enlarge]

That's the luma waveform from the Color Tools filter in VirtualDub. If you don't know how to read a waveform this might help:

https://forum.videohelp.com/threads/340804-colorspace-conversation-elaboration#post2121568

Basically, the waveform graph transforms the brightness of the pixels to a height on the graph. With a horizontal graph like this the graph is an overlay of the brightness of all the pixels above that part of the graph. You can see that the darkest parts of the picture are, for the most part, at the left and right edges. Closer to the middle are the boy's black pants. But the are nowhere near the desired black level. That could be because they aren't fully black under the lighting conditions, or because the black level is too high. It's up to you to decide where you want the black level to be. Here I've moved the black level down to just above that dark peak at the left is crushed (all 0).


[Attachment 66875 - Click to enlarge]

[SkyBlues2021]

Thanks Sharc. I will try both.

[SkyBlues2021]

Thanks jagabo. I can see the colours look more natural and less washed out on the right hand photo in your test. I've read the waveform link, however I will read it a few more times hopefully it will sink in. I will try it on a clip and I may understand it better.
Would those filter levels be used throughout a 1 hour video or is this a scene by scene adjustment.

[lollo]

You can also use the AviSynth function Limiter(), to highlight the area where the levels are outside the range 16-235, and then decide if they need to be considered and tuning the levels while capturing or not, as Sharc properly suggested:

function f_limiter(clip c)
{
c_l=c.Limiter(16, 235, 16, 240, "luma_grey")
c_c=c.Limiter(16, 235, 16, 240, "chroma_grey")
stackhorizontal(\
c,\
stackvertical(\
c_l.BicubicResize(c_l.width/2,c_l.height/2),\
c_c.BicubicResize(c_c.width/2,c_c.height/2)\
)\
)
}

[Attachment 66878 - Click to enlarge]


[Attachment 66879 - Click to enlarge]

[Sharc]

Would those filter levels be used throughout a 1 hour video or is this a scene by scene adjustment.

The answer is - as often - it depends.
For systematic level shifts like elevated blacks the same correction can usually be applied throughout.
In other cases like changes of the light conditions or backlit shots, a scene by scene adjustment may be necessary. There exist algos in Avisynth which make such adjustmens dynamically, e.g. AutoLevels, AutoAdjust or Retinex (see my example in post#18). Sometimes with unwanted side effects like flicker. One has to try and find the compromise.
See also the discussion here:
https://forum.videohelp.com/threads/406406-Restoring-compensating-overexposed-heavily-...eather-video-s

[jagabo]

You can also use the AviSynth function Limiter()

Yes, that is another good tool.

I hesitate to mention this but even within those limits (Y = 16 to 235, U and V = 16 to 240) most YUV combinations lead to illegal RGB values. YUV is a rotation and scaling of the RGB value:


https://www.intel.com/content/www/us/en/develop/documentation/ipp-dev-reference/top/vo...-models_FIG6-5

The inner cube is all 16 million 8-bit per channel RGB triads. The outer cube is all 16 million 8-bit YUV triads. Only about 1/6 of all YUV combinations lead to valid RGB values. Even within the limited range only about 1/4 of the YUV combinations lead to valid RGB values.

If you're a real stickler for valid values you need to do a lot more than check for Y=16 to 235, and UV=16-240. One quick dirty method is to convert the YUV image to RGB then back to YUV. Any YUV triads outside the RGB cube will have their RGB values clamped within the valid RGB range. Then on conversion back to YUV the values will be different than the original YUV image. I created a filter long ago that highlights bad YUV values using only built in AviSynth filters:

https://forum.videohelp.com/threads/360935-Capturing-Correct-Chroma-and-Hue-Levels-Fro...e2#post2289672

Also in that thread are some other methods. Notable are some that use Gavino's more accurate (but much slower) method.

[lollo]

Yes jagabo, thanks, I have seen your previous posts about this topic.

At the capture side, however, there is really not much we can do, just stay inside the range of the capture card.

[Sharc]

At the capture side, however, there is really not much we can do, just stay inside the range of the capture card.

Hmmm... thinking about it: If it would be possible to show the RGB Parade (avisynth: HistogramRGBParade(coeffs="601")) we could check whether the R,G,B waveforms are clipped at 0 or 255) which would indicate "illegal" Y,Cb,Cr (means YUV triplets outside the inner cube) suggesting proc-amp adjustments. Just a thought...... or a fart

[SkyBlues2021]

Thank you for the advice. I've done some new capture which I'm hoping are not as washed out. I adjusted the proc amp and the colours seem richer. Any advice would be appreciated. Thanks.

[SkyBlues2021]

I missed this one.

[lollo]

Hmmm... thinking about it: If it would be possible to show the RGB Parade (avisynth: HistogramRGBParade(coeffs="601")) we could check whether the R,G,B waveforms are clipped at 0 or 255) which would indicate "illegal" Y,Cb,Cr (means YUV triplets outside the inner cube) suggesting proc-amp adjustments.

Sure, technically is possible, but I would not stress the procamp (and the user) to capture inside these limits, but only inside the capture card range limit. It is already difficult doing the second for a whole tape, and will deteriorate too much the whole capture just to catch some triads out of range

[Sharc]

Thank you for the advice. I've done some new capture which I'm hoping are not as washed out. I adjusted the proc amp and the colours seem richer. Any advice would be appreciated. Thanks.

I had a brief look only and yeah, it's getting gradually better with the capture levels (dark levels). Ready for final post-processing I think.

Sidenote: With clip 20-9-22.avi you are approaching the issue as described in post #39, e.g. with the red component (in RGB) around frame#156. No need for action though. It's not unusual for VHS tapes, mostly for the red and blue component.

[SkyBlues2021]

Thanks Sharc. I moved everything towards the left hand side and only had the odd dip into the red area, I will adjust it slightly before I do a full 1 hour capture. The colours looked a lot more coloured. But I'm hoping It's getting something like a acceptable standard now?

[Sharc]

Hmmm... thinking about it: If it would be possible to show the RGB Parade (avisynth: HistogramRGBParade(coeffs="601")) we could check whether the R,G,B waveforms are clipped at 0 or 255) which would indicate "illegal" Y,Cb,Cr (means YUV triplets outside the inner cube) suggesting proc-amp adjustments.

Sure, technically is possible, but I would not stress the procamp (and the user) to capture inside these limits, but only inside the capture card range limit. It is already difficult doing the second for a whole tape, and will deteriorate too much the whole capture just to catch some triads out of range

Basically agree. In practice it's normally both, means appropriate capture levels settings according to the capture device' range limits, and final post-processing.
The RGB clipping test would mainly serve as an indication. When the so called "proc-amp" adjustments are made in the digital realm anyway (typical for USB capture devices) it would do no harm to achieve fully "legal" captures (or as much as we accept a compromise) in one single step rather than in 2 steps, because we would have the integer rounding effects (combed histogram with gaps and quatizing errors) happening only once. Anyway, this just crossed my mind.

[jagabo]

I brought up the YUV/RGB details just for a little more completeness. I didn't mean one should worry too much about it while capturing.

I only looked a a couple of the latest caps but the black level is still to high, and the white level too low.

[SkyBlues2021]

Thanks jagabo. I do think these captures look better and less washed out than my previous captures. How would I combat the high black level and low white level with the internal proc amp? I'm guessing less brightness and less contrast?

[jagabo]

Less brightness and more contrast.

In most programs "brightness" adds or subtracts a value to the luma. For example, if a pixel has a luma value of 1 and you add 1 it becomes 2. If a pixel has a luma value of 100 and you add 1 it becomes 101. "contrast" is a multiplier. With a contrast value of 2.0 a pixel with a value of 1 becomes 2. A pixel with 100 becomes 200. In some programs these are called offset and gain (because that's what they really are at the analog hardware level).

Using the Brightness/Contrast filter in VirtualDub:

[Attachment 66909 - Click to enlarge]

Here I used Levels to adjust the video. Because Levels also has a gamma adjustment that brings out shadow detail. I let the black bars get crushed. And the white text a bit too.


[Attachment 66905 - Click to enlarge]

[SkyBlues2021]

I've attached 3 new clips. I've tried to reduce the contrast as best I can whilst staying within the safe zone. clip (2) has slightly different settings to clip (3) & clip (4) thanks.

[SkyBlues2021]

Thanks jagabo. I didn't see your post before posting. I will do some more clips with less brightness and more contrast.

[jagabo]

Once gain, your being fooled by the black bars at the left and right. Those aren't part of the active picture. You don't care whether they get crushed or not. That's why a histogram is near useless.

But your white levels are getting better.

[SkyBlues2021]

Thanks jagabo. I did look at the waveform but maybe wrongly I thought it was post capture? I know Sharc's earlier post mentioned the black bars fooling me and I thought that he didn't think I was cropping the borders & swishing out before looking at the histogram. However now you're saying the same thing maybe I misunderstood.
If I don't use the histogram how should I come at this and stay within the 16-235 range?

[SkyBlues2021]

I've just been looking at the waveform. I noticed on the avipmod waveform that there is void between the bottom brown bar and the white wave. The top appears to be closer. I'm assuming the wave should be much closer?
The virtualdub waveform seemed fairly good, so I imagine I'm probably not reading it correctly. I also uploaded my virtualdub settings for this.
Is this a method of checking the 16-235 range as well as the histogram.
Looking at the Avipmod waveform I must get my capture closer to the bottom bar? Thanks.

[jagabo]

The black bars in your Color Tools waveform is very similar to those inthe AviSynth waveform. The dashed line is Y=16 in Color Tools.

[SkyBlues2021]

Thanks jagabo. Just thinking should I have removed the borders before doing the waveform?

[Sharc]

.......Just thinking should I have removed the borders before doing the waveform?

Not necessarily. If you leave the black borders you may even use them as a reference for black. In post#55 (AvsPmod) they should be at 16, now they are at about 30. Reduce the black level accordingly.
If you crop them off you have to spot a part of the active picture which you think is "nearly black" and set the black level such that the waveform is close to 16 (say 18...20) for that part of the picture.

[SkyBlues2021]

Thanks Sharc. I will post some more clips tomorrow. Hopefully when I reduce the black levels, I will get closer to an acceptable capture.
Just a thought how do most folk do this, do they scan through a tape and extract clips and analyse them with the waveform before capturing to set the required levels?

[jagabo]

Yes, in theory those black bars should be at Y=16. But a lot of VHS recordings have blacks higher than that (ie, nothing in the active picture ever gets as dark as the black bars). So you may need to set the proc amp so that those bars are lower than Y=16. As Sharc points out, a more reliable method is to search through your tape for the darkest darks you can find (over a large area, not just oversharpening halos), something you think should be black, and use that as a black reference for capturing.