Using a color-reduction
reference might not be the best option, as it hides some subtle but important volume information, for instance the nose highlights. All the more since the light on the face is very soft.
I understand the rationale of reduxing refs to help visualize shading, except in my experience it doesn't really work: you have little control over the cut-off values so the resulting shapes are arbitrary (and often ugly); also there are complex causes behind values in a photograph and not all are related to the main light source(s), eyes can discriminate better.
If you're going to do it, compare results from different redux settings (color counts) and only pick global information such as comparison between broad areas.