Automatic cameras capture each of the three color components on a scale of 0 to 255. They calculate the exposure so that the average is 128 for the three components. Depending on the contrast of the scene, one or more of the colors may fall outside the capability of the camera, so that a component value that should be greater than 255 will be limited to 255.

Suppose that the subject is entirely shades of red. We would like the middle value of the red subject to be captured as 128, with the other shades of red arrayed above and below that. However, the automatic exposure will set (M + 0 + 0)/3 = 128. The green and blue components are zero, and the autoexposure sets the average of the three components to 128. Instead of M being set to 128, M/3 is set to 128. The image is overexposed by one and a half stops.

Few subjects are dominated by one color component, so autoexposure usually works well. For example, blue sky appears blue, but actually there is good deal of white mixed with the blue. Flowers are an exception. Flowers can have very pure colors. White flowers or pastel shades do not pose a problem. Usually it is red or red-orange flowers that create a problem. Some green foliage and green vegetables are also problematic.

Here is an image of a poinsettia taken with autoexposure. (The red things are actually leaves, but, hey, play along.)

I have superimposed histograms of R + G + B and of the red channel alone on the image. This is a high contrast situation, so it is no surprise that the histogram looks truncated at both the high and low ends. However, when the red channel is viewed, it is apparent that many of the shades of red were so overexposed as to fall off the high end of the scale.

Here is a series of images with the expose starting at the automatic level and then decreased in one-quarter stop increments:

The last image is two stops underexposed with respect to the autoexposure. In the higher exposures the intense red tones are transformed to white or near white as a result of the red being limited to 255 while the minority green and blue components are boosted.

The red histogram shows values centered below 128 — roughly a half stop under. One and a half stops agrees well with the theory. The contrast was so high that even with two stops underexposure there are still some values truncated at 255. That leads to some blocked highlights, but nonetheless a much better image than the autoexposure.

I tweaked the two-stops-under image by increasing the shadows by 6% and decreasing the highlights by 7%. This effectively tempers some of the contrast. The final image is then:

The rule of thumb for subjects having pure colors is to force the exposure to one and a half stops below the automatic determination.