RGB can NOT describe colour

[Deleted member 16349]

It has long been known theoretically that one can not get accurate colour values from RGB values. That has not stopped developers in the graphic arts to imply they can.

An interesting blog post by John Seymour of QuadTech covers this issue and provides interesting tests to highlight this problem. He also talks about his years dealing with this problem and his final conclusions. It is clear he spent a lot of time and frustrating effort with this issue, which frankly could have been avoided if just understanding and accepting that theoretically it was not so workable.

His interesting blog is at this site.

John the Math Guy: RGB into Lab

But what does this mean? It means that RGB workflows can not accurately describe colour. Many developers might claim that it is good enough but this is foolish. The reason it is foolish is because when one tries to solve a problem with an approach that conflicts with known theory, one can not know the magnitude and location of the errors that will result.

For years I have discussed the problem issues in the press which cause the lack of consistency and predictability. There has been no real interest to get these issues fixed. The RGB problem is just as important and I expect there will be no interest to get it fixed either.

From the recent rant by Frank Romano about colour management and other related issues on an Enfocus webinar, he clearly showed his frustration but of course he had no solutions. His frustration stems from the lack of action by the industry to understand the core issues and properly address them.

The industry lives in two worlds. One where publications, associations and suppliers state how well things are being done with new technology and the other world where every day problems continue to stress people due to the lack of consistency and predictability.

Systemic problems are the worst because they will constantly be a source of frustration and waste.

 

[gordo]

While the post is technically correct it somewhat misses the point. Image reproduction is seldom about absolute accuracy measured numerically but accuracy in the sense of maintaining the perceived appearance of the subject.
It is very much a subjective endeavour.
The reason for RGB images and workflows is not about accuracy of reproducing colors. It is about flexibility in image purposing.
In print reproduction, accuracy usually means aligning presswork with a reference - the proof. That alignment is always conditional - e.g. Change the lighting and the color will likely no longer align.
What developers in the graphic arts are trying to do is establish specific aim points under specific conditions for aspects of the imaging process. Their purpose is to bring some level of color alignment across different print facilities and technologies - when that alignment is required (it not always is).
In that context, accuracy is about hitting the target aim point numerically within the specified tolerance. i.e. Not subjectively. Given the vagaries of measuring instrument and the wide tolerances for the aim points - achieving the goal of aligning across different print facilities and technologies remains problematic.
At least having these specifications in place brings some consistency to the process, and as you well know, consistency is more important than accuracy.

 

[Deleted member 16349]

While the post is technically correct it somewhat misses the point. Image reproduction is seldom about absolute accuracy measured numerically but accuracy in the sense of maintaining the perceived appearance of the subject.
It is very much a subjective endeavour.
The reason for RGB images and workflows is not about accuracy of reproducing colors. It is about flexibility in image purposing.
In print reproduction, accuracy usually means aligning presswork with a reference - the proof. That alignment is always conditional - e.g. Change the lighting and the color will likely no longer align.

I don't think RGB can maintain a perceived appearance if it can not predict a resulting colour even if the printed result is viewed under standard controlled lighting. That is just my view. RGB is just not capable. Period.

RGB is maybe easy to use, but for who. There should have been attempts to find other ways to solve these basic problems that would be easy to use and result in reasonably accurate results. Absolute accuracy is not the goal.

John Seymour's experiments do show results in different lighting, which are interesting but really has nothing to do with the fact that one can not go from RGB to colour values with any predictable accuracy.

 

[gordo]

I don't think RGB can maintain a perceived appearance if it can not predict a resulting colour even if the printed result is viewed under standard controlled lighting. That is just my view. RGB is just not capable. Period.

RGB is maybe easy to use, but for who. There should have been attempts to find other ways to solve these basic problems that would be easy to use and result in reasonably accurate results. Absolute accuracy is not the goal.

John Seymour's experiments do show results in different lighting, which are interesting but really has nothing to do with the fact that one can not go from RGB to colour values with any predictable accuracy.

Well it may be a moot point anyway because RGB doesn't actually have color values. Instead it contains bitmap data - effectively tone values.
RGB doesn't describe color.
If you want values that represent a color with predictable values then you could use CIEL*a*b* combined with a LUT as used in a certain vendor's spot color simulation system.

 

[Deleted member 16349]

Yup, something like that is the future in my mind. Not CIEL*a*b* but yes to LUTs. There are problems with using Lab values when calculating colour in an image at different scales.

 

[Stephen Marsh]

My take on the aim of the article was that digital capture for colorimetric measurement purposes is not a good idea for accurate results in graphic arts applications. So no, an iOS or Android OS app and smart phone hardware does not make a good spectrophotometer, and no, a video or still camera on a press is not great tool for colour measurement accuracy.

This is of course very different to taking this article out of context and saying that non-spectral based colour capture or RGB processing or RGB workflows have no place in photography or graphic arts.

Most camera sensors are monochrome, using a “Bayer” or similar colour filter array to interpolate RGB (or other) colour values from the raw photon intensity values captured by the sensor (known as demosaicing). Less commonly one may find a “Foveon x3” R, G, B layered sensor or for product photography a “scanning back camera” that has three separate R, G, B passes.

As the popularity of editing colour in Photoshop attests, most human observers are not after “accurate” colour from RGB captures, they are after “pleasing” colour (granted this is a subjective term and there is no baseline). There have been studies performed and it has been found that for certain key “memory colours” most human observers find “unrealistic” colour more pleasing than “realistic” colours (sky and foliage are two common examples in landscape photography).

Both RGB and CMYK values are just that - numbers, we need to have a description of what type of RGB or CMYK those values mean, which is generally in the form of an ICC profile. Once we have both a value and an associated profile, we can then have a known L*a*b* value.


Stephen Marsh

 

[Deleted member 16349]

Both RGB and CMYK values are just that - numbers, we need to have a description of what type of RGB or CMYK those values mean, which is generally in the form of an ICC profile. Once we have both a value and an associated profile, we can then have a known L*a*b* value.


Stephen Marsh

I don't think what you have said is valid in any general way. But what you have said still supports the comment that RGB data can not describe colour on its own.

Maybe RGB with a profile for a specific known process can provide a reasonable colour value but how do you get the profile. You have to measure colour coming out of the process. So the colour information comes from a measurement at some point used in the profile and not from the RGB values.

 

[Stephen Marsh]

I don’t think what you have said is valid in any general way.

? :]

But what you have said still supports the comment that RGB data can not describe colour on its own.

Nor can CMYK. Both are device dependent colour models. 255r0g0b or 0c100my0k are both sets of values/numbers. They have a “meaning” in that they are both as “pure/saturated” a red as can be achieved in each colour model - but beyond this they don’t mean anything from a colorimetric viewpoint. Assign/assume a newsprint profile against 100m100y and one has very different Lab colour numbers than when one assumes a sheet fed profile against the same 100m100y values.

The article in question was referring to the problems of using RGB based sensors to measure colour (please correct me if I am wrong in the specifics of the focus of the article).

Erik, my impression is that you are using the conclusions of the article out of context, however I could be wrong there as well, so please take the time to flesh out you reply so that I can understand what your “point” is.


Stephen Marsh

 

[Deleted member 16349]

? :]



Nor can CMYK. Both are device dependent colour models.

The article in question was referring to the problems of using RGB based sensors to measure colour (please correct me if I am wrong in the specifics of the focus of the article).

Erik, my impression is that you are using the conclusions of the article out of context, however I could be wrong there as well, so please take the time to flesh out you reply so that I can understand what your “point” is.


Stephen Marsh

Device independent colour data is required. So yes, CMYK can not describe colour.

I am not making my conclusions from the article. I have known about this for over ten years. I learned this from reading related papers and discussions from colour scientists on newsgroups.

Yes , John Seymour is concerned with capturing colour and has come to the conclusion after a long time that it is not possible with RGB devices for reasonable results. That is why QuadTech has gone to a spectral capture technology for their in press colour control system.

My point is clear. RGB can not describe colour. If one uses RGB to deal with colour, it will not be so predictable.

People may still feel that it is OK to deal with RGB for colour management but that is their problem and the problem of the industry. I am just pointing out a problem and included John's article for interest. Different technical approaches are required if there is to be easier and more predictable methods to manage colour but most likely they will not be developed.

 

[JohnTheMathGuy]

Thank you, Stephen for clarifying my blog post. I only demonstrated that RGB sensors of any sort don't make good color measurement devices. This goes for iPhone apps, inexpensive RGB color measurement devices, high-end studio cameras, and industrial cameras used to monitor printing inline. The blog post should not be construed to reveal anything about any other part of the workflow.

But... Since RGB cameras are used as part of the workflow, the inaccuracy of color measurement does come into play. A agree, Stephen, that images are meant to be pleasing and not accurate... except in catalog work. There, the images must be both pleasing (to sell the goods) and also accurate (to avoid having it returned)!

 

[Stephen Marsh]

John, thank you for joining the discussion.

If you feel that I have “clarified” your blog post, then I am honoured ;] I was just trying to establish the context of Erik’s use of your blog post as reference material, as I was a little confused as to where Erik was coming from and going to with his statements.

Would you prefer comments specific to the blog post to go against your blog or here?


Stephen Marsh

 

[JohnTheMathGuy]

Stephen, please feel free to put comments wherever it's convenient. I get comments on the blog, on LinkedIn, Facebook, and Twitter, wherever people see my links. Often I get private messages. If there are some interesting comments, I share them in a followup blog post.

 

[rich apollo]

If you change the light, you change the color. This is fact.

Doesn't John's experiment show that if you change the observer, you change the color? What I mean is, doesn't this show that John's eyes and the cameras he tested have very different sensor responses?

 

[JohnTheMathGuy]

Yup... my eyes have different spectral response than the camera that I used... and different from every RGB sensor that I have seen. That is the crux of the problem.

(I'm guessing that your eyes are probably different from the camera's as well.)

 

[Deleted member 16349]

Yup... my eyes have different spectral response than the camera that I used... and different from every RGB sensor that I have seen. That is the crux of the problem.

(I'm guessing that your eyes are probably different from the camera's as well.)

I would say that if you had a camera which has the same spectral response as your eyes, you still would not get the right colour values. You would need a camera that had a response that was the same as the x,y and z colour matching functions, which are not the same as the spectral response of the eyes.

That is the way I understand it. Also the math seems to define this.

 

[D Ink Man]

Anyone have the algebra or mathematic formula for calculating multiple L*a*b* values of multiple colors to target an exacting L*a*b* value?

Hope this belongs here, seems the right minds are present.

Thank you in advance. D Ink Man

 

[Deleted member 16349]

Anyone have the algebra or mathematic formula for calculating multiple L*a*b* values of multiple colors to target an exacting L*a*b* value?

Hope this belongs here, seems the right minds are present.

Thank you in advance. D Ink Man

D, not exactly sure what you are asking but if you are asking for some formula to take several Lab values, sum and average them to get a single Lab value, then I have to say that it can not be done accurately.

You can do it with the multiple XYZ tristimulus values to get an accurate total XYZ tristumulus value, which you can then convert to Lab. That will then result in an accurate Lab value.

 

[D Ink Man]

OK, let me further complicate the matter which will actually better explain it.

Using pigmented colorant bases, (like pMS) I arrive at a Lab value with X amount of ingredients. It can be anywhere from 2 to as many as 6 bases which of course include t-white and neutral black. These two bases by the way are big players and movers of Lab value, worth noting. Let's say I have 12 different reds with up and down varying values for Lab and base percentage variations. I am looking for the mathematical formula to provide an exacting Lab value using any combination of the 12 different reds to provide a targeted Delta E of 'zero'. Once I have the mathematical problem calculated to provide this, I can simply bring the proper combination to percentage and have a blend formula to match my predetermined Lab value. It sounds complicated, but it really isn't with the proper formula. The answer is out there, just need the equation. It is a trilogy with the Lab movers, but it can be done in my sought application.

To do it without a computer program would even be more satisfying, but that will do if available. Ty D

 

[Deleted member 16349]

OK, let me further complicate the matter which will actually better explain it.

Using pigmented colorant bases, (like pMS) I arrive at a Lab value with X amount of ingredients. It can be anywhere from 2 to as many as 6 bases which of course include t-white and neutral black. These two bases by the way are big players and movers of Lab value, worth noting. Let's say I have 12 different reds with up and down varying values for Lab and base percentage variations. I am looking for the mathematical formula to provide an exacting Lab value using any combination of the 12 different reds to provide a targeted Delta E of 'zero'. Once I have the mathematical problem calculated to provide this, I can simply bring the proper combination to percentage and have a blend formula to match my predetermined Lab value. It sounds complicated, but it really isn't with the proper formula. The answer is out there, just need the equation. It is a trilogy with the Lab movers, but it can be done in my sought application.

To do it without a computer program would even be more satisfying, but that will do if available. Ty D

D, maybe you are thinking it could be something like this.

Target Lab = Lab1 x k1 + Lab2 x k2 + Lab3 x k3 + ..... where k1 + k2 + k3 + ... are fractions of 1 and all add up to 1 for a mixture of a given number of base inks. (Note: L, a, b calculated separately)

The above equation might look OK but it is not. It will not be accurate.

Sorry if this does not help. I can't tell for sure what you would like. I do know some ink suppliers have formulas to try to predict colours but they tend to be approximate and require a bit more adjusting.

 

[Schnitzel]

John, regarding inline industrial cameras: if the sensor is working in a controlled environment, its spectral response is known and an ICC profile can be generated for it, doesn't that make it predictable and suitable for color measurement?