A very useful way to really get a handle on this null is to zoom in on the picture. I've used Reinhard's Face pattern here for a reason!
The reason is that as you stay within a single column, scan downward until the "shadow" side of a face appears to be "in shade". That's the left side of each picture. In the right, treat it as though the coloured side is "lighting" the "face".
At some point as you traverse the vertical rows, you'll realize that this emergent cognition *changes*.









