Assuming that some object in the scene (field of view ) approximates "white" and independently normalizing each channel to set the maximum value to "white" is another method of spectral normalization commonly used.

Here are the results of "normalization to white", applied to the test images.

Two dimensional histograms of the corrected images are available: chromaticity and Red vs. Blue

Color Corrected Test Images

Gamut	Illuminant A	Illuminant B	Illuminant C
All
Little Blue
Little Red

These results are abysmal ! It turns out that the simple implementation used doesn't work very well when there are specular reflections in the scene. I modified the algorithm to correctly implement the white world assumption, and tried again

For quick reference, here are the test images (All gamut) :

Illuminant A	Illuminant B	Illuminant C