[Tig] Illuminance Curves

Dan C. Tatut dtatut
Sun Jan 22 12:22:10 GMT 2006

Thanks to oktobor for supporting the TIG
I can only agree with you.... What I was trying to say is exactly that: terms like luma/chroma luminance/chrominance depend on the context. By that I meant that the available litterature on the subject is not totally unified. Also terms used in the color correction machines/systems/software are not always accurate.
Your "illuminance"-based techniques are great and you can definitely do whatever you want. My point was just that as long as we speak about creative color correction, anything is possible/permitted.
I know Charles Poynton's papers very well and the only thing I can say is that there is enough controversy on the web about them, I will not add my 2 cents here. Also color corrections based on definitions like Rec 709 are valid (again it's my opinion) as long as the whole system remains coherent. In the case of the color correction hardware boxes I'm talking about, what happens is that the video signal is modified (corrected/adjusted) with the purpose of displaying it on a video (SD/HD) device. These boxes work in a closed environment (video-to-video).
When you start using software and software/hardware correctors, then the story is different. The Rec 709 does not mean anything on a computer monitor because the chromaticity primaries are not the same as for a video monitor. Because of this, what the software does is definitely purely theoretical.
Beware when working with Photoshop. It is far from being perfect!!!! Photoshop uses a 3D LUT (color cube) to transform the RGB in LAB and vice versa. The transformation represented by the color cube depends on input and output color management profiles. If your Photoshop is not correctly parameterized, then the results are far away from anything useful. Also because of the technique they use, the tranformation is subject to interpolation (loss of precision) and distortions and clipping (the color cube actually remaps colors to match some medium, it is not only a straight transformation). Because of this, you will never have accurate results. 
I suggest you have a look at this web site: http://brucelindbloom.com
It might help a lot here. Discovering how these hardware boxes function is not difficult, I agree. The reason I would not spend time doing that is because replicating their algorithms is not something that will help colorists. There are much better ways for color correcting video and film material than working with these boxes. If you have a chance to attend NAB, I'll give you some live, visual examples.... very convincing.
The fact that colorists should know how the algorithms work, I totally agree. Actually it is the only way to select the best tool to achieve a result. Take Photosop as an example. I can easily think of 5/6 different tools to achieve the same correction. Which one is the best? It's necessary to understand how they work, their advantages and disadvantages. Too often I've seen people using the wrong tool and wasting hours to achieve the desired effect.... scary.
What you say about Final Cut, is the very reason why colorists need color correction applications. If you look at the DI-emerging market, the reason for this poliferation of tools is exactly that. Colorists have been "boxed" within these famous boxes for too long. It's about time to unleash them...
Best, Dan

Dan Tatut 
CHROME Imaging 
105 Rue de Lyon 
CH-1203 Geneva 

Phone: +41 22 807 23 60 
Fax: +41 22 807 23 70 
Mobile: +41 78 659 11 04 
WWW: http://www.chrome-imaging.com <http://www.chrome-imaging.com/>  

-----Original Message-----
From: glenn chan [mailto:glennchan at gmail.com]
Sent: samedi, 21. janvier 2006 23:01
To: Dan C. Tatut; tig at tig.colorist.org
Subject: Re: [Tig] Illuminance Curves

Hi Dan,

Thanks for the reply.

1- My original argument was that I don't know of any color correction systems that implement "illuminance" curves (illuminance as in whichever of R, G, or B is greatest).  The maximum of R, G, and B is distinctly different than some combination of R, G, and B.

2- I want to check to see if we're on the same page.  To me, luma and luminance are very different things!  Similarly, chroma is different from chrominance.

Example 1:
Assume that if you change the iris on a camera, this only affects luminance.

A vectorscope measures chroma, and should not be affected by luma information at all.

If you sit with a camera hooked up to a vectorscope and adjust its iris, you see that the vectorscope trace changes.  This is because luminance is not the same as luma.  Chroma is proportional to luma levels.  Hence lower exposure means lower chroma levels if chrominance is to remain constant.

There is also a subtler difference between luma and luminance.  See http://www.poynton.com/papers/YUV_and_luminance_harmful.html

Example 2:
The Color Corrector filter in Final Cut Pro has sliders that adjust luma, not luminance.  So if you take the "whites" slider (equivalent to gain?), you need to adjust the chroma control (labelled, confusingly in my opinion, as "saturation") to keep saturation constant.

3- If the ratio between R:G and G:B are constant, then hue and saturation are constant.

Multiplying R, G, and B by the same number changes luminance.  If you think of light in the physical domain this makes sense.  Moving a light closer or futher away from a reflective object will not change the ratio between R:G:B.

I think we're in agreement here.  Before I wrongly assumed that LAB would keep the ratios between R:G:B constant.  I just checked in Photoshop, it doesn't seem to keep these ratios constant.

4- Suppose you were to define luminance as some combination of R, G, and B.  In my opinion. the best-looking algorithm would use the formula:
"Luminance" (I use this term very loosely here!) = ( R + G + B ) / 3.

Compare this to rec. 709:
Rec. 709: Luma (Y') = 0.2126 R' + 0.7152 G' + 0.0722 B'

Pure blue is seen as a really dark color compared to pure green.  So you have something 'bizarre' happening where greens become brighter, and blues become darker.  By averaging R, G, and B instead all the three primary colors are instead seen as the same brightness.  However you still get an effect where neutral colors are seen as brighter than saturated colors.

5- I don't think it would be that hard to figure out the formula someone is using in their system.  With an editing machine + SDI in/out, you can send and capture processed test patterns.  From there you should be able to determine what formulas they are using (visually or numerically).  You can easily and legitimately guess what algorithms a system uses.

I think colorists should know what algorithms a system uses as it plays a factor in output quality.  For example, there are many ways to de-saturate an image (in Sony Vegas, there are 3/4).  As the TIG article on contrast shows, for simple things like contrast there are a variety of algorithms that can be used.

Some algorithms in my opinion should generally be avoided.  i.e. luma and chroma based algorithms are useful only in niche situations (i.e. video levels errors).  Final Cut Pro's 3-way CC in my opinion is really annoying since it is based on luma and chroma. 

Tig list - http://tig.oktobor.com/mailman/listinfo/tig
TIG wiki: http://tig.colorist.org/wiki

More information about the Tig mailing list