The Color Parrot custom white balance attachment

Drew Strickland, proprietor of the Pro Photo Home forums, has introduced a new "custom white balance attachment" called the Color Parrot. You can read about it here:

http://www.prophotohome.com/forum/p...argeted-repeatable-digital-white-balance.html

The Color Parrot is placed in front of the camera lens and the camera aimed at the subject. A "white balance reference frame" is then taken (under the camera's "custom white balance" scheme). The camera's analysis of that frame is then used to establish a color correction vector to be applied in-camera to the actual shot data.

The Color Parrot is said to provide superior results to other similar tools owing to properties I can't really identify.

More importantly, I continue to me mystified as to how this approach can consistently work (with any attachment). Our need is to ascertain the chromaticity of the illumination on the subject. The inverse of the departure of that chromaticity from that of the "reference white" for our color space is then the "color correction vector" that needs to be applied to the taken image.

If we have an instrument (a colorimeter, or a camera equipped with a "white balance acceptance diffuser"), placed at the subject location, with the face of the instrument parallel to the subject surface, it will accept the incident illumination just as the subject will, and its determination of the chromaticity of that light is just what we need.

But if the "instrument" is placed at the location from which the shot will be taken, and aimed at the scene, the light that it gathers is the reflected light from the subject. The chromaticity of that light (from any given "spot" in the scene) is the joint result of the chromaticity of the incident illumination on the spot and the reflective chromaticity of that spot. The light accepted by the "operative region" of the camera's white balance reference frame will be some averaging of that light over the acceptance angle of the instrument, weighted in some way over various arrival angles.

The only way that the light captured by the "instrument" will assuredly have the chromaticity of the illumination is if the average reflective chromaticity of the scene (averaged under the angular weighting of the instrument) is "neutral".

Of course, we sometimes arrange for just that by placing in the scene a reflective chromaticity-neutral target (a "gray card") and take our reference frame (without any lens attachment) with only that target in the operative region of the frame.

But when the instrument is aimed at the scene "as is", I just cannot see how it can consistently discern the needed fact: the chromaticity of the illumination.

My attempts to have the distributor of the Color Parrot explain how it acquires the needed information under its mode of use have been rebuffed ("This has been discussed ad nauseam"). I am assured that the color balance results produced by this attachment are really good. So there. Kumbaya.

Perhaps I am missing something. If so, can anyone here tell me what it is?

Thanks.

Doug_Kerr said:

If we have an instrument (a colorimeter, or a camera equipped with a "white balance acceptance diffuser"), placed at the subject location, with the face of the instrument parallel to the subject surface, it will accept the incident illumination just as the subject will, and its determination of the chromaticity of that light is just what we need.

Click to expand...

Yep, that's the goal and it can only be achieved accurately by measuring the spectral mix of incident and ambient light incident on the subject. This implies that Drew's product may be sub-optimal in achieving accuracy, but maybe good enough when used from a distance pointing at the subject. Whether it's better than AWB may depend on the particular AWB implementation.

The only part of the concept that looks beneficial to me, if I were shooting JPEGs (which I very rarely do), is creating a Custom WB that is applied to subsequent shots. However, the only thing the color parrot does is take a weighted average reflected color image, and hope the camera's AWB does a better job on that than on a detailed image (where the latter potentially can do much better with a good algorithm).

But if the "instrument" is placed at the location from which the shot will be taken, and aimed at the scene, the light that it gathers is the reflected light from the subject. The chromaticity of that light (from any given "spot" in the scene) is the joint result of the chromaticity of the incident illumination on the spot and the reflective chromaticity of that spot. The light accepted by the "operative region" of the camera's white balance reference frame will be some averaging of that light over the acceptance angle of the instrument, weighted in some way over various arrival angles.

Click to expand...

Exactly, and the question then becomes: Does the camera's Custom WB determination produce a better result than the AWB based on a more detailled rendering of the scene? Whether it does, and AWB should actually be better because it has more clues to take into consideration, depends on the quality of the AWB algorithm.

My attempts to have the distributor of the Color Parrot explain how it acquires the needed information under its mode of use have been rebuffed ("This has been discussed ad nauseam"). I am assured that the color balance results produced by this attachment are really good. So there. Kumbaya.

Perhaps I am missing something. If so, can anyone here tell me what it is?

Click to expand...

You're not missing anything, it's just that the truth can hurt his commercial success.

The remaining question thus becomes: Does such a Custom WB based on a (weighted) average scene color provide a better approximation than a CWB from e.g. a piece of paper with unknown amounts of brightener?

Bart

I don't think you're missing anything. That "white balance acceptance diffuser" is a hardware version of "auto white balance" and when using it one is assuming the incident mixed with the reflective and diffused light will be a balanced gray result.

That works pretty well for "average" subjects and when the color balance (or viewer) isn't critical. From what I can tell, most people don't care as long as people in the shot aren't green. Yellow is ok until it becomes so pronounced a diagnosis of liver failure is warranted.

Hi, Will,

Will_Perlis said:

I don't think you're missing anything. That "white balance acceptance diffuser" is a hardware version of "auto white balance" . .

Click to expand...

Actually, I think not. I have always assumed that the AWB process uses some "intelligent" algorithm.

. . . and when using it one is assuming the incident mixed with the reflective and diffused light will be a balanced gray result.

That works pretty well for "average" subjects and when the color balance (or viewer) isn't critical. From what I can tell, most people don't care as long as people in the shot aren't green. Yellow is ok until it becomes so pronounced a diagnosis of liver failure is warranted.

Click to expand...

Indeed.

Thanks for chiming in.

Best regards,

Doug

Hi, Bart,

Bart_van_der_Wolf said:

However, the only thing the color parrot does is take a weighted average reflected color image, and hope the camera's AWB does a better job on that than on a detailed image (where the latter potentially can do much better with a good algorithm).

Click to expand...

Well, one uses the Color Parrot to take a reference frame for a custom WB determination, and the AWB algorithm is not used in evaluating that frame (CWB and AWB are two entirely different processes.).

. . .and the question then becomes: Does the camera's Custom WB determination produce a better result than the AWB based on a more detailled rendering of the scene? Whether it does, and AWB should actually be better because it has more clues to take into consideration, depends on the quality of the AWB algorithm.

Click to expand...

It's hard to believe that a "mindless" examination of the scene (by CWB, through any non-magical attachment) should produce consistently better results than an "intelligent" examination of the scene (by AWB).

Best regards,

Doug

"I have always assumed that the AWB process uses some "intelligent" algorithm."

I'm sure it does, but it *has* to be making some assumptions about the mix it is looking at if it can't see the ambient light. The hardware version is doing much the same by the choices made of the amount of diffusion, the plastic's color neutrality, the angle of acceptance, etc. No?

Hi, Will,

Will_Perlis said:

"I have always assumed that the AWB process uses some "intelligent" algorithm."

I'm sure it does, but it *has* to be making some assumptions about the mix it is looking at if it can't see the ambient light. The hardware version is doing much the same by the choices made of the amount of diffusion, the plastic's color neutrality, the angle of acceptance, etc. No?

Click to expand...

Yes, but the hardware version can't vary its "decisions" based on what it sees, whereas the AWB algorithm presumably does (that is, it is "adaptive").

Best regards,

Doug

Well, I was wrong

Earlier today I expressed some misgivings as to how a color parrot could be used in making a white balance determination.

Well, I was wrong, and now I get it. It can be very useful.



Best regards,

Doug

Doug_Kerr said:

Earlier today I expressed some misgivings as to how a color parrot could be used in making a white balance determination.

Well, I was wrong, and now I get it. It can be very useful.

Click to expand...

LOL, no, ROFLMAO.

I have to get myself one of those assistants.

Bart

Doug,

Thanks for the laugh and for showing me my monitor is still calibrated nicely for neutral gray.

Will

I missed the answer

Well, I had earlier said that the distributor of the Color Parrot did not provide any explanation of how it manages to determine the chromaticity of the ambient light when the camera-plus-Parrot is aimed at the subject.

I reread the thread a little more patiently, and it turns out that it has been explained. So my apologies.

Here's the explanation:

The fact is shooting the subject [aiming the camera-plus-Parrot at the subject for the reference frame] works well most of the time. This is not the product of "luck." [I had said that making a measurement by aiming at the subject would only work through luck.] It is the result of the fact that when light strikes an object it reflects light back. By passing this reflected light through a white target to create a reference image the camera is able to create a reasonably accurate set of numbers that can be used to create a reasonably accurate suggested white balance preset.

Well, who'd a thunk it! How did I miss that.

Best regards,

Doug

Will_Perlis said:

Doug,

Thanks for the laugh and for showing me my monitor is still calibrated nicely for neutral gray.

Will

Click to expand...

Doug, you need to realize that some of us get these threads via email (sans images). When I saw what you wrote, I was wondering what was going on. Glad I came to the site to see your excellent gray balanced work! Now I realize it was done tongue in cheek (thankfully).

Hi, Andrew,

Thanks for writing.

What is your take on the method used by the Color Parrot (not using my colored parrot)?

Best regards,

Doug

Doug_Kerr said:

What is your take on the method used by the Color Parrot (not using my colored parrot)?

Click to expand...

Lets just say I'm skeptical. Certainly for those of us shooting Raw.

Doug_Kerr said:

Yes, but the hardware version can't vary its "decisions" based on what it sees, whereas the AWB algorithm presumably does (that is, it is "adaptive").

Click to expand...

Hi Doug,

Correct, that's the difference between what's known as the 'gray world' versus 'true' white balancing algorithms. True AWB is most likely implemented in a color space that separates Chromaticity from Luminance (e.g. Lab). In addition, bright (=desaturated) 'colors' are deemed to be closer to real neutral (e.g. high L, and low 'ab'), and certain 'colors' (e.g. blue sky, green grass, human skin, etc.) are excluded from the equations. The resultant (masked) average image color is likely to be subjected (after conversion to XYZ color space) to 'Robertson's algorithm', to derive the main black-body temperature involved (which leaves a 'tint' variable to quantify). Of course, everything depends on the implementation of an 'intelligent' AWB algorithm, versus a more naive 'Gray world' algorithm.

Bart