WB measurement with flash - Part 1

Simplistically, white balance color correction is the process of transforming the originally recorded colors in an image to overcome the phenomenon in which the colors of scene objects are not "correctly" perceived by viewers of the delivered image because the original photograph was taken under ambient light whose chromaticity did not match the white point of the color space in which eh image was recorded.

In order to do this, the correction "mechanism" must know the chromaticity of the incident light upon the subject.

In my opinion, in the general case there is only one foolproof way to know that: measure it. And, in my opinion, the only foolproof way to measure it is where it is (which is at the subject location).

There are two common methods:

1. Use an instrument (perhaps a color photometer, or perhaps our camera itself equipped with an appropriate "acceptance diffuser"), placed at the subject location with its accepting surface oriented in an appropriate direction.

2. Place temporarily in the scene a "target" whose reflective chromaticity is "neutral" (such as a "gray card") and have the camera regard it, then or later ascertaining the chromaticity of the ought reflected from the target.

Both of these are invasive of the scone, and as such may be impossible, impractical, socially unacceptable, or just a pain. Thus we would all rather have a way to do what we need to from the comfort and relative convenience of out camera position.

By how, by observing something from that location, can we ascertain the chromaticity of the light incident on the subject (which is someplace else)? Well, most modern digital cameras will try and do that for us from the taken image itself, through their "automatic white balance" mode. In this, artificial intelligence is used to attempt to divine what, if any, areas in the scene are likely "neutral" in their reflectance, or have a non-neutral reflective chromaticity that can be guessed at. The results of this are highly variable, and we can rarely say that they satisfy our aspirations in most of the cases of interest.

Some workers believe that we can somehow, from the shooting position, have an "instrument" (again, perhaps our camera equipped with an acceptance diffuser) observe the scene, and from a "mass" measurement of the light reflected from it, discern the chromaticity of the incident light. These aspirations are almost never accompanied by any scenario as how this can actually be expected to work reliably.

The dilemma, of course, is that the light reflected from scone objects is a creature of both the chromaticity of the ambient light (which we would like to know) and the reflective color of the subject area (which we also want to know).

Drew Strickland, the manufacturer of the Color Right white balance measurement diffuser (sometimes known as the ColorRight, and previously known as the Color Parrot, says that this diffuser, by virtue of its rather narrow "acceptance pattern", is better able to overcome this dilemma than other diffusers, which typically have a "broader" acceptance pattern. No story as to how this happens has been advanced.

Recently (in fact, in part at Drew's suggestion), I made a couple of tests that can shed some light on this situation. The results interlock well with the results of a complementary series of tests recently disclosed by Drew himself on the Color Right Sales and Support section of the Pro Photo Home forms, collectively supporting my outlook on what we should expect in different situations.

My tests (in order to somewhat parallel Drew's) were in a studio-like setting in which all the illumination was by electronic flash. Basically, a humanoid subject was photographed, with white balance color correction applied "in camera" (using the Canon "custom white balance" facility), with the "reference frame" taken in three ways:

1. By having the camera observe a fairly-large neutral target (specifically, a Picture Perfect gray card)

2. By having the camera regard the actual scene, equipped with a "wide pattern" white balance measurement acceptance diffuser (namely, an ExpoDisc diffuser, 2007 model).

3. By having the camera regard the actual scene, from the shooting position, equipped with a "narrow pattern" white balance measurement acceptance diffuser (namely, a Color Parrot diffuser, Version 1.2).

Note that the inclusion of the ExpoDisc diffuser in the series was done somewhat against my better judgement. Its manufacturer does not recommend it for making white balance measurements "from the shooting position". Nevertheless, in his recent test reports, Drew tested both with the ExpoDisc (in the "from the shooting position" mode and the Color Right (in that same mode).

I pointed out that this comparison was not really meaningful, since the ExpoDisc was not intended for that form of combat. To the contrary, said Drew, it was a fair comparison; he reported that the performance of the ExpoDisc in his tests was less accurate than the Color Right, but the fact that the ExpoDisc was not recommended for that mode does not let it off the hook.

Rather, he says, that of itself is a point of superiority of the Color Right, since the "from the shooting position" mode is often the only practical one, or at least by far the most convenient. (This line of logic has a kink in it somewhere, but I have been urged by some of my colleagues not to obsess over what I find to be inconsistencies in Drew's pronouncements.)

In any case, thus provoked, I included a test with the ExpoDisc in this series.

[continued in Part 2]

WB measurement with flash - Part 2

[continued]

Now to my tests. The subject was a manikin named Stella. She is 5'9" tall as presently configured, and wore a honey blond wig for the shoot. She was accessorized with a pedigreed WhiBal gray card, worn off-the-shoulder. The purpose of this was to allow after-the-shot determination of how closely the color correction approached the "theoretical ideal", in which a neutral subject in the scene would be recorded as "reference white" in the image color space (here, sRGB).

The background was a process blue Botero fabric backdrop, approximately 8' x 8' in size.

The camera was a Canon EOS 20D, using a Canon EF 24-105mm f/4L IS lens. The only consequential light source was a Canon 580EX II flash unit in the camera shoe. The camera was operated in M mode to hold the shutter speed and aperture constant. The flash unit was operated in M mode to keep the output constant (and thus to avoid the small change in output chromaticity than comes with change in output level). The beamwidth (head zoom) of the flash unit was fixed at its widest setting (to assure that the backdrop, a player in this drama, was fully illuminated.

As I mentioned above, three white balance refernce frames were taken. In the ones with the diffusers, the exposure was "bumped" (by increase in aperture) for the reference exposure to assure adequate photometric exposure for chromaticity determination by the camera.

Then, three actual shots of Stella were taken, one with each of the three frames as the reference for in-camera custom white balance color correction.

We see the three resulting images here. They have been cropped to a handy scope and are presented here at reduced resolution. However, no other adjustments have been made to the original images.

Balanced with a reference frame taken by observation of a Picture Perfect gray card



Balanced with a reference frame taken with an ExpoDisc (2007) diffuser, measuring from the shooting position.



Balanced with a reference frame taken with a Color Parrot (v1.2) diffuser, measuring from the shooting position.




For each image, I determined the RGB coordinates averaged over a small square area in the gray card. These were converted to offsets from the white point in terms of the CIE u'v' chromaticity diagram. Those offsets are shown here:



Note the following:

1. We did not attempt to take into account the factory-specified calibrated chromaticity of the WhiBal gray card. It was treated as if perfectly neutral.

2. We note that in reading the chromaticity of the gray card in the image, the coordinates were read on an 8-bit RGB basis, and so there is an exposure to quantizing error that can approach 0.01 u'v' unit.

[Continued in Part 3]

WB measurement with flash - Part 3

[continued]

Before I comment on the significance of these results, let me mention the complementary tests reported by Drew Strickland, with which my tests nicely interlock. They differed from my tests principally in the following regards:

1. The background behind the subject was said to be white (presumably close to neutral)

2. The illumination was partly by flash and partly by something else.

3. There was no test with a gray card.

The results of examination of the resulting images are shown here:



Now, what do these results tell me? Well, there is really no surprise at all.

Drew's tests

First, in Drew's tests, the backdrop was neutral. The white balance measurement with the Color Right diffuser primarily embraced light from the backdrop (owing to its relatively-narrow acceptance pattern). Thus the light we would expect the diffuser to take in, preponderantly from reflection from an essentially-neutral surface, would indeed closely represent the chromaticity of the incident light. (This is in effect a "giant gray card" measurement.) Thus we would expect a rather accurate white balance color correction in this case.

The ExpoDisc (remember, not intended for this type of measurement at all) would probably pick up light from beyond the span of the backdrop - possibly by reflection from a bluish wall. Thus of course it would make a chromaticity reading that differed from the incident light, and the color correction guided by that measurement would err (in this case, in the red-yellow direction).

My tests

In my tests, the backdrop was blue. The white balance measurement with the Color Right diffuser primarily embraced light from the backdrop (owing to its relatively-narrow acceptance pattern). Thus the light we would expect the diffuser to take in, preponderantly from reflection from a blue surface, would not closely represent the chromaticity of the incident light, but would depart in the blue direction. (This is in effect a "giant blue card" measurement.) As a result, we would expect the color correction guided by that measurement to have an error in the yellow direction.

The ExpoDisc (remember, still not intended for this type of measurement at all) would probably pick up light from beyond the span of the backdrop - in this case, likely by reflection from a beige wall (the color is actually called "Moose Mousse"). Thus of course it would make a chromaticity reading that differed from the incident light in the red direction, and the color correction guided by that measurement would err in the cyan direction.

And, son-of-a gun, that's exactly what happened, across the board.

But enough of the theory. Which of these results were nice pictures? All of them.

WB mearement with flash - The missing link

I had neglected in my discussion of my WB measurement test in a flash setting and those of Drew Strickland to give a link to his report.

Here it is:

http://www.prophotohome.com/forum/c...ight-works-better-flash-than-competition.html

Best regards,

Doug