Doug Kerr

April 15th, 2008, 06:18 AM

The Canon EOS cameras have a "white balance shift" setting. This provides for setting a "color correction" of up to ±9 units along two axes, designated "amber-blue" and "green-magenta".

I thought it would be interesting to see just what these two axes are and how they relate to the axes of known color spaces.

I tested by having my EOS 20D regard a WhiBal gray card exposed to a consistent ambient illumination. I used the CWB function to get a basic WB correction for the illumination. Then I took test exposures with the WB shift set to all nine possible combinations of -5, 0, and +5 units on each of the two axes (treating the "amber" direction and "green" directions as positive, the convention used in reporting the WB shift setting in the Exif metadata).

I plotted the results on several different coordinate planes, along with vectors showing the "directions" of the chromaticity of the three color space primaries (R, G, and B) and the complementary hues C, Y, and M. The result for the "0,0" setting was used as the reference, arbitrarily plotted at the origin of the coordinate system.

The results on the CIE u-v plane were the most telling. There, as a matter of fact, the six "primary" vectors are almost equally-spaced angularly.

Here we see that plot. The axes are labeled "du" and "dv" to reflect that this is a "differential" plot; the actual origin of the u-v plane is far off this chart.

http://Pumpkin.Annex.home.att.net/figures/20d_wb_shift_uv.jpg

Note that since I read the resulting chromaticities of the color-corrected images on an 8-bit RGB basis, the plot is susceptible to errors from quantizing that can approach 0.001 u-v unit.

The light violet lines connect the extreme points of the constellation of data points, a primitive way to estimate the orientation of the two axes without benefit of formal linear regression analysis.

The two black double-ended arrows show the orientations of these two axes.

We see that the two axes are not quite orthogonal (at right angles) on this plane. The "green-magenta" axis is seen to be closely aligned with the green and magenta primary vectors. The "amber-blue" axis falls between two pairs of primary vectors; thus we see that the labels "amber" and "blue" are somewhat arbitrary. (The two directions are actually an "orange" hue and a "cyan-blue" hue.)

On the larger front, we note that evidently the chromaticity shift produced by this setting "piles onto" the shift established by the "other" white balance setting, regardless of the choice. That is, whatever color shift would be produced by the current WB setting (including a decision made by the camera's AWB algorithm), the "WB shift" adds to that.

The Canon manuals generally indicate that, with respect to settings along the amber-blue axis, one unit of shift corresponds to 5 mireds of color temperature correction.

The mired is a unit of inverse color temperature (now preferably called the "mirek", or even more preferably, the "micro reciprocal Kelvin")

It properly only quantifies chromaticity shift along the "blackbody locus", that is. among the chromaticities produced by a black body radiator at various temperatures. The "amber-blue" axis of the Canon WB shift lies generally in that direction. (Since the locus is a curve on the u-v plane, no axis can truly follow it along its length.)

Similarly, the green-magenta axis, being orthogonal to the amber-blue axis on the CIE u-v plane, is generally the direction along which we measure the departure of a chromaticity from the blackbody locus (the direction reflected by the "tint" parameter in Photoshop).

I thought it would be interesting to see just what these two axes are and how they relate to the axes of known color spaces.

I tested by having my EOS 20D regard a WhiBal gray card exposed to a consistent ambient illumination. I used the CWB function to get a basic WB correction for the illumination. Then I took test exposures with the WB shift set to all nine possible combinations of -5, 0, and +5 units on each of the two axes (treating the "amber" direction and "green" directions as positive, the convention used in reporting the WB shift setting in the Exif metadata).

I plotted the results on several different coordinate planes, along with vectors showing the "directions" of the chromaticity of the three color space primaries (R, G, and B) and the complementary hues C, Y, and M. The result for the "0,0" setting was used as the reference, arbitrarily plotted at the origin of the coordinate system.

The results on the CIE u-v plane were the most telling. There, as a matter of fact, the six "primary" vectors are almost equally-spaced angularly.

Here we see that plot. The axes are labeled "du" and "dv" to reflect that this is a "differential" plot; the actual origin of the u-v plane is far off this chart.

http://Pumpkin.Annex.home.att.net/figures/20d_wb_shift_uv.jpg

Note that since I read the resulting chromaticities of the color-corrected images on an 8-bit RGB basis, the plot is susceptible to errors from quantizing that can approach 0.001 u-v unit.

The light violet lines connect the extreme points of the constellation of data points, a primitive way to estimate the orientation of the two axes without benefit of formal linear regression analysis.

The two black double-ended arrows show the orientations of these two axes.

We see that the two axes are not quite orthogonal (at right angles) on this plane. The "green-magenta" axis is seen to be closely aligned with the green and magenta primary vectors. The "amber-blue" axis falls between two pairs of primary vectors; thus we see that the labels "amber" and "blue" are somewhat arbitrary. (The two directions are actually an "orange" hue and a "cyan-blue" hue.)

On the larger front, we note that evidently the chromaticity shift produced by this setting "piles onto" the shift established by the "other" white balance setting, regardless of the choice. That is, whatever color shift would be produced by the current WB setting (including a decision made by the camera's AWB algorithm), the "WB shift" adds to that.

The Canon manuals generally indicate that, with respect to settings along the amber-blue axis, one unit of shift corresponds to 5 mireds of color temperature correction.

The mired is a unit of inverse color temperature (now preferably called the "mirek", or even more preferably, the "micro reciprocal Kelvin")

It properly only quantifies chromaticity shift along the "blackbody locus", that is. among the chromaticities produced by a black body radiator at various temperatures. The "amber-blue" axis of the Canon WB shift lies generally in that direction. (Since the locus is a curve on the u-v plane, no axis can truly follow it along its length.)

Similarly, the green-magenta axis, being orthogonal to the amber-blue axis on the CIE u-v plane, is generally the direction along which we measure the departure of a chromaticity from the blackbody locus (the direction reflected by the "tint" parameter in Photoshop).