Can anyone explain Colorspace and LUTs?
by Loren White
on
Dec 23, 2008 at 8:45:49 am
I can't find a clear definition on these two things anywhere.
My current idea of ColorSpace (REC709, RedSpace, etc etc)
is basically a color profile or setting for the colors that are applied to the reference broadcast monitor. This ensures that the colorist grades with the same color space as he is outputting for.. HDTV, NTSC, Digital Cinema, etc?
Can anyone elaborate on that?
My current idea of LUTs is almost the same thing. It is a Look Up Table, (which I see as a color setting) for the image, basically a preset grade, that gives the image a look from which to start grading off of?
Re: Can anyone explain Colorspace and LUTs? by Joseph Owens on Dec 23, 2008 at 5:41:37 pm
A LUT is a map, can be 2D or 3D that re-plots values between colourspaces. Colourspaces only exist because presentation media have different capabilities. Computer monitors are not grade monitors are not consumer TVs are not print magazines are not projection theatres.
Gamut is the issue. Cross-reproduction media do not use the same center values as the sources, and indeed are not even capable of creating the desired rendition. You can generate numbers in one system that are "fish on a bicycle" to another. For example, just for simplicity consider the extremely simple flip from YCbCr to RGB and back. The math doesn't work. Anyone who has simply made the non-corrected roundtrip trip with FCP and COLOR knows that. One is additive-only and the other is a blend that uses subtractive matrixed colour difference. YCbCr is capable of generating negative values.
Consider film. Yes, real Original Camera Negative that requires printing to a positive. The colour image there resides on CMYK layers. Print stocks are matched to the neg's characteristics to put the "key in the lock" so that the result corresponds to what the eye would, under specific illuminant conditions.
To create a specific colour value -- hue, saturation, brightness, we need three numbers -- but these can also be relative, now thinking about "white point" colour temperature. Returning to our original camera negative, or even the CCD image sensors on a video camera, these values are generated by filtering systems -- the layers of the OCN itself in film, and dichroic mirrored prisms or (some other voodoo like the Bayer pattern in the RED). So who chooses the filter colours? And what exactly do they generate? Three images, (in most cases) a red one, a green one, and a blue one -- that when recombined will yield a recognizable "real world" picture. And now a word about wavelengths and frequencies. So what if the red filter/image on your originating media doesn't exactly match the "red", whatever that may be, on your reproduction device -- say, a slightly different variant of the return filtering in a scan/telecine, or more to the point, the phosphor/LCD illuminant of a video monitor? How do your colour components recombine? Slightly wrong. So that shade of canary yellow (and I'm thinking specifically of Warner Brothers' "Tweety Bird") isn't going to be right. Anybody else here have any sense of how tightly controlled an animated character like that is? "Rabid" would be a good word, at least in my experience, but it does go beyond that.
So, in short, LUTs are for trying to map values into different colourspaces to achieve the same visual impresssion.
Re: Can anyone explain Colorspace and LUTs? by Loren White on Dec 24, 2008 at 2:57:45 am
Thanks for the info.
Any real world examples to guide me?
So, Apple Color is in RGB color space, while FCP is in YCbCr. Yet they are being displayed on the same computer style monitor DVI/HDMI? Why would they use different color spaces?
So a LUT could allow us to match our footage to what it looked like in FCP (YCbCr LUT)? Giving us a pre-compensating grade? Does a LUT affect the entire monitor or just the image in the UI viewer? Where would LUTS be applied? (Ive never had to deal with them, though I should).
Re: Can anyone explain Colorspace and LUTs? by Jeremy Garchow on Dec 24, 2008 at 6:35:35 am
[Loren White]"So a LUT could allow us to match our footage to what it looked like in FCP (YCbCr LUT)? Giving us a pre-compensating grade? Does a LUT affect the entire monitor or just the image in the UI viewer? Where would LUTS be applied? (Ive never had to deal with them, though I should)."
I know this has been said already, but you have to do some reading.
open Color, go to the Help menu and choose Color User Manual. Then in the search bubble type in LUT
Re: Can anyone explain Colorspace and LUTs? by Joseph Owens on Dec 24, 2008 at 4:15:48 pm
LUTs shouldn't be required for separate viewing of RGB / YCbCr material, since they should both adhere to the 601/709 colourspace, gamma adjusted for SD or HD. There can be significant differences between displays depending on whether FCP is compensating for a DVI computer monitor profile and the output on a grade monitor showing the HDSDI output of a Kona card -- but that's not a broadcast problem, its a personal-computer problem. Its still not a LUT issue, especially not a 3D situation. If you need a LUT to get calibrated/characterized color on a computer monitor... that's a bigger problem, and dangerous in practice because you will never be looking at a true rendition -- always a simulation. Like trying to print a metallic finish on an inkjet. They are just different worlds.
The real problem with switching between RGB and YCbCr is that the round-trip math is approximate and sometimes doesn't work at all, especially if there are negative values involved. Take split-field color bars as a "for instance". There was a recent thread on the TIG regarding 'pluge', the little 5/7.5/10 IRE black reference signal. In 601, the analog 7.5 pedestal black is now "0" in the digital 601 world. What do you do with the "5" IRE level? Is it minus-something? No, in reality video 0-100 is sampled (in 8-bit) to correspond to 13-235 in the 0-255 world, allowing for some over- and under-shoot. But what if it goes "deeper" than that? Red and blue chromaticity is especially famous for generating negative composite values. The "I" and "Q" quadrature flags (the bluish and magenta-ish squares either side of the peak white square) are based on 0-luminance and +/- 20 IRE of saturation. They are way out of gamut in RGB and are impossible to represent even in 0-255. No problem in YCbCr, though, because it is a subtractive representation, and chromaticity is mapped around a center value, so you can have tons of saturation at both ends of the luminance scale.
All of this is poison for video, especially with upper and lower clipping, which will return false balance.
Which is why the "auto" button is not your friend, and LUTs don't cure everything.
As to my problems:
From what I see the FCP LUT implementation is totally wrong; the .mga format from Pandora is actually a 1D LUT only - while FCP uses it as a 3D Cube; the cube also uses 16 points, not 17 like every other system that I've worked with...
And to cap this it seems that FCP has a restricted colour space, so a full RGB cube, even with 16 points, will show errors when the colour space goes beyond the limited one of FCP, as any RGB cube will...
Real bad, and makes me think FCP may actually using YUV colour internally???
The demo 3D Cube builder that can be downloaded from the link above has the ability to build FCP 3D LUTs, and from it you can see some of the issues (the cube is called FinalTouch for historical reasons)...
The black dots in the image viewed through the LUT are a watermark - the rest of the image should be correct, except for the issues FCP has - which seems to make the image 'green' overall if the cube goes too far... I'd appreciate it if anyone can have a play and let me know what they think.
