Removing color artifacts without using greyscale

[bruno321]

If I'm encoding some material shot in black and white, usually greyscale() does a good job of removing those annoying color artifacts. But in this case I feel greyscale() is changing the tint too much: Original color,

[Attachment 55607 - Click to enlarge]

Greyscaled,

[Attachment 55608 - Click to enlarge]

I know it's subjective, but I feel the latter has become bluish. How can I keep the same black and white feel of the original while making sure all those color artifacts are gone, and without changing the tint too much?

[darkknight145]

If you open these images in a image editor and use the eye dropper you'll see that the original is too red in the mid greys (white and black are okay). The greyscaled image is perfect across all levels. The only way you might be able to do this in a good video editor where you might be able to add back some tint.

[Atak_Snajpera]

If Grey scale looks bluish then IT means that you have badly calibrated monitor. (Too cold color temperature set)


http://www.lagom.nl/lcd-test/

[butterw]

Your original is sepia color, it's not clear what you mean by color artifacts. A grayscale function removes tint (by nulling YUV chroma, or by applying a custom LumaCoeff in RGB), but you can reapply tint if you wish and as you wish.
https://forum.doom9.org/showthread.php?p=1925869#post1925869

[butterw]

Interestingly your grayscale image has a pattern of periodic holes in the histogram, maybe try a different grayscale function that avoids this issue.

[jagabo]

You can make the image greyscale then add back the chroma offsets. That won't be prefect because the offsets vary by intensity but it might be close enough for you. ColorYUV(cont_u=-256, cont_v=-256).ColorYUV(off_u=-3, off_v=3). You could get a close match buy using masks to apply different offsets to different intensities.

You didn't provide a sample where there are rainbow artifacts but you might be able to blur them away leaving the overall colors intact with something like MergeChroma(last, BinomialBlur(5)).

As usual, a video sample which shows the artifacts would be more useful.

[bruno321]

Thank you all for the comments. Here's a picture of the color artifacts:



and a short snippet: https://www.sendspace.com/file/7llsdy

Jagabo's ColorYUV line has very good effects (I don't know what it does precisely; I'll try to understand it from the manual pages but if you want to help with a tl;dr I won't complain!). The MergeChroma one reduces the rainbows but doesn't get rid of them entirely.

EDIT, I know think I understand what those two lines do. ColorYUV(cont_u=-256, cont_v=-256) seems to be the same as greyscale(), and ColorYUV(off_u=-3, off_v=3) concentrates the U and V more or less where the spikes of the original image are (saw this using histogram("levels")). Makes sense!


Here's a soft question: this is from a film shot in 1965 with an Arriflex 16 ST https://en.wikipedia.org/wiki/Arriflex_16ST . I want to keep the look closest to the original film (not the DVD transfer). Can I know whether I should be keeping the sepia-ish tint (as in the DVD) or if I should be greyscaling?

[smrpix]

Here's a soft question: this is from a film shot in 1965 with an Arriflex 16 ST https://en.wikipedia.org/wiki/Arriflex_16ST . I want to keep the look closest to the original film (not the DVD transfer). Can I know whether I should be keeping the sepia-ish tint (as in the DVD) or if I should be greyscaling?

The camera had no effect on the color of the film. It should be black and white to match the original negative or, more likely, reversal stock (though there are some white dust specs suggesting an internegative.) The DVD transfer has far greater artifacts -- improper density, scratches -- suggesting the slight sepia tint was not deliberate. (Loved using those Arris.)

[bruno321]

Great, a reply from somebody who used those cameras! Much appreciated I know very little about how those old black and white cameras worked, but I suspected they are just chemically incapable of capturing color.

So I guess I should just be using greyscale() indeed. The thread was very informative, at any rate.

[manono]

So I guess I should just be using greyscale() indeed.

I do it with a Tweak(Sat=0,Coring=False). Same thing.

[jagabo]

EDIT, I know think I understand what those two lines do. ColorYUV(cont_u=-256, cont_v=-256) seems to be the same as greyscale(), and ColorYUV(off_u=-3, off_v=3) concentrates the U and V more or less where the spikes of the original image are (saw this using histogram("levels")). Makes sense!

Yes, exactly.

The MergeChroma one reduces the rainbows but doesn't get rid of them entirely.

I forgot that BinomialBlur() doesn't blur the chroma by default. You can use this to blur only the chroma channels (no need for MergeChroma())

BinomialBlur(Y=2, varc=10.0, U=3, V=3)

The rainbows still aren't completely removed, but their much improved.

[smrpix]

Great, a reply from somebody who used those cameras! Much appreciated I know very little about how those old black and white cameras worked, but I suspected they are just chemically incapable of capturing color.

So I guess I should just be using greyscale() indeed. The thread was very informative, at any rate.

Arri-s cameras are perfectly capable of capturing color. It's the film that's determinative.

[bruno321]

Ah, seems I misunderstood what you wrote. So is there a possibility this "not entire black and white" is how it was in the original film?