Doug Kerr
Well-known member
As many of you know, we do not support the notion that a "reliable" determination of incident illumination chromaticity (as needed for white balance color correction) can be made by a measurement made from the camera position (that is, the camera position, and aiming, to be used for the actual shot), regardless of what "instrument" is used (excluding "artificial intelligence" instruments, such as a camera's "automatic white balance" facility).
But now we have a "white balance measurement diffuser" said by its manufacturer to produce reliable results for that measurement technique, the Color Parrot version 1.2.
So we thought we might want to give it a try in actual use.
By way of background, the manufacturer tells us that the Color Parrot's abilities in that regard are largely conferred by its relatively-narrow "directivity pattern". As with this technique generally, we cannot understand how that property can enable the technique to function reliably. But we were certainly willing to look into that claim.
We did not take the large collection of shots that would be needed to gave a broad view of the success of this technique. Were we planning to do so, we would probably want to take and analyze perhaps five shots each taken under perhaps ten clearly-distinct illumination and shooting situations. We leave that to others of a more-masochistic bent.
Rather, we took shots using this white balance measurement technique in a single photographic setting: a human model, shot with fairly tight framing, in an illustrative "indoor lighting" situation. The camera used was a Canon EOS 40D. The lens used was a Canon EF 24-105 mm f/4.0 IS USM lens.
We also took the same shot using white balance correction based on a WhiBal gray card target at the subject.
In each case, the result of the "white balance measurement" was used to set a "custom white balance" (CWB) refernce in the camera, which was then used for in-camera white balance color correction of the various actual shots.
The three white balance measurement techniques used were:
a. WhiBal gray card target, held by the subject at chest height, measured as recommended by the camera manufacturer.
b. Color Parrot 1.2, used with the camera in the precise location and aiming situation used for the shot proper. (The lens was arbitrarily set to its shortest focal length and the focus to infinity.)
In the actual shots, the model still held in place the WhiBal gray card. We have credible data on its reflective chromaticity. Thus, by measurement of its "reflected light" chromaticity as recorded by the camera, with in-camera color correction applied, compare to its "known" reflective chromaticity, we can quantitatively judge the "accuracy" of the overall white balance color correction.
That chromaticity was measured in two ways: by examination of the JPEG output image in an editor, reading the chromaticity on an 8-bit RGB basis, averaged over a small rectangular area in the center of the image of the gray card target. (Given the fairly large deviations encountered her, we did not feel it was worth the trouble to go through he 16-bit TIFF conversion to read the chromaticities on a 16-bit basis. Thus we must recognize a source of experimental error from quantizing error. We estimate that error to be on the order of 0.001 du'v' unit.
We then compared that chromaticity with that which would have been ideally recorded for the gray target (that is, had white balance color correction been achieved in the "theoretically ideal" way).
Some may question the significance of the result for a white balance refernce measurement using the WhiBal gray card, since measurements on that same target were used to assess the "accuracy" of the end result. But since we "know" the reflective chromaticity of the gray card (from the manufacturer's data), then the determination of overall white balance performance, even with this device playing its dual role, is valid. (I'll spare you the algebraic proof.)
If of course one wants to say, "Well, that data for the WhiBal gray card was not independently verified" and thus the entire procedure is suspect", then I suggest you ignore this entire report and await a comparable report from a laboratory with fancier reference sources than we have.
But. moving right along. . . .
Here we see the delivered, color-corrected JPEG images for both of the color balance measurement techniques. The offset of the chromaticity of image of the gray target from its "ideal" value is noted in terms of a du'v' value.
Color corrected based on a reference frame taken from a WhiBal gray card at the subject position (same position shown in the shot). Total discrepancy from "perfect correction": 0.0011 d u'v' unit (in the cyan direction).
Color corrected based on a reference frame taken with a Color Parrot 1.2 diffuser "from the camera position"). Total discrepancy from "perfect correction": 0.053 d u'v' unit (in the cyan direction).
(You can just tell how excited Carla is about the ongoing parrotry!)
Finally, here we show that offset from "theoretically ideal color correction" in each color-corrected shot, plotted on the u'-v' plane:
Comment
We are not surprised by the direction of the discrepancy of the color correction with the Color Parrot ("from the camera position"). There was a lot of red surface within its "view", and the direction of the discrepancy is consistent with our expectations for that.
Still, the correction there is not outrageously far from the "theoretical ideal". We attribute that to the fact that, even with the fairly narrow directivity pattern of the diffuser, there were still a number of objects in its scope that were probably, on average, "not that far from neutral".
The narrow pattern of the device makes it come closer to fulfilling the "reflected light" description often given (inappropriately, for most diffusers) to the "from the camera position" measurement technique. As I sad before, if that's what one thinks is desirable, then we have it here.
Certainly either of those images would have been "acceptable" in a number of contexts. We've seen bluer skin than that in Color Parrot demo pictures.
We wish all users of "from the camera position" white balance measurement (with whatever tool) similar good fortune.
But now we have a "white balance measurement diffuser" said by its manufacturer to produce reliable results for that measurement technique, the Color Parrot version 1.2.
So we thought we might want to give it a try in actual use.
By way of background, the manufacturer tells us that the Color Parrot's abilities in that regard are largely conferred by its relatively-narrow "directivity pattern". As with this technique generally, we cannot understand how that property can enable the technique to function reliably. But we were certainly willing to look into that claim.
We did not take the large collection of shots that would be needed to gave a broad view of the success of this technique. Were we planning to do so, we would probably want to take and analyze perhaps five shots each taken under perhaps ten clearly-distinct illumination and shooting situations. We leave that to others of a more-masochistic bent.
Rather, we took shots using this white balance measurement technique in a single photographic setting: a human model, shot with fairly tight framing, in an illustrative "indoor lighting" situation. The camera used was a Canon EOS 40D. The lens used was a Canon EF 24-105 mm f/4.0 IS USM lens.
We also took the same shot using white balance correction based on a WhiBal gray card target at the subject.
In each case, the result of the "white balance measurement" was used to set a "custom white balance" (CWB) refernce in the camera, which was then used for in-camera white balance color correction of the various actual shots.
The three white balance measurement techniques used were:
a. WhiBal gray card target, held by the subject at chest height, measured as recommended by the camera manufacturer.
b. Color Parrot 1.2, used with the camera in the precise location and aiming situation used for the shot proper. (The lens was arbitrarily set to its shortest focal length and the focus to infinity.)
In the actual shots, the model still held in place the WhiBal gray card. We have credible data on its reflective chromaticity. Thus, by measurement of its "reflected light" chromaticity as recorded by the camera, with in-camera color correction applied, compare to its "known" reflective chromaticity, we can quantitatively judge the "accuracy" of the overall white balance color correction.
That chromaticity was measured in two ways: by examination of the JPEG output image in an editor, reading the chromaticity on an 8-bit RGB basis, averaged over a small rectangular area in the center of the image of the gray card target. (Given the fairly large deviations encountered her, we did not feel it was worth the trouble to go through he 16-bit TIFF conversion to read the chromaticities on a 16-bit basis. Thus we must recognize a source of experimental error from quantizing error. We estimate that error to be on the order of 0.001 du'v' unit.
We then compared that chromaticity with that which would have been ideally recorded for the gray target (that is, had white balance color correction been achieved in the "theoretically ideal" way).
Some may question the significance of the result for a white balance refernce measurement using the WhiBal gray card, since measurements on that same target were used to assess the "accuracy" of the end result. But since we "know" the reflective chromaticity of the gray card (from the manufacturer's data), then the determination of overall white balance performance, even with this device playing its dual role, is valid. (I'll spare you the algebraic proof.)
If of course one wants to say, "Well, that data for the WhiBal gray card was not independently verified" and thus the entire procedure is suspect", then I suggest you ignore this entire report and await a comparable report from a laboratory with fancier reference sources than we have.
But. moving right along. . . .
Here we see the delivered, color-corrected JPEG images for both of the color balance measurement techniques. The offset of the chromaticity of image of the gray target from its "ideal" value is noted in terms of a du'v' value.
Color corrected based on a reference frame taken from a WhiBal gray card at the subject position (same position shown in the shot). Total discrepancy from "perfect correction": 0.0011 d u'v' unit (in the cyan direction).
Color corrected based on a reference frame taken with a Color Parrot 1.2 diffuser "from the camera position"). Total discrepancy from "perfect correction": 0.053 d u'v' unit (in the cyan direction).
(You can just tell how excited Carla is about the ongoing parrotry!)
Finally, here we show that offset from "theoretically ideal color correction" in each color-corrected shot, plotted on the u'-v' plane:
Comment
We are not surprised by the direction of the discrepancy of the color correction with the Color Parrot ("from the camera position"). There was a lot of red surface within its "view", and the direction of the discrepancy is consistent with our expectations for that.
Still, the correction there is not outrageously far from the "theoretical ideal". We attribute that to the fact that, even with the fairly narrow directivity pattern of the diffuser, there were still a number of objects in its scope that were probably, on average, "not that far from neutral".
The narrow pattern of the device makes it come closer to fulfilling the "reflected light" description often given (inappropriately, for most diffusers) to the "from the camera position" measurement technique. As I sad before, if that's what one thinks is desirable, then we have it here.
Certainly either of those images would have been "acceptable" in a number of contexts. We've seen bluer skin than that in Color Parrot demo pictures.
We wish all users of "from the camera position" white balance measurement (with whatever tool) similar good fortune.