Here are a few demonstrations showing that shadows have to be darker — although this constraint is checked only at the shadow borders. The first is a video of Stuart Anstis taken in our lab where he is shown in black and white and in outline. As you will see, certain objects he holds up are easy to interpret in their outline versions whereas others, typically those depending on discriminating shadow from object, are not

Stuart and the Fly

Next is a set of two-tone faces which required the shadow areas to be properly identified before the 3D structure of the face can be recovered. The outer two start in negative contrast where the shadow areas are lighter and the faces unrecognizable. The middle face may be just slightly in positive contrast with shadow areas appropriately darker, and so recognizable. As the video runs, the red gets darker until the outer two faces switch to positive contrast where they should be recognizable, whereas the middle one becomes negative (so not recognizable). The luminance then fades back to the original values.

Two-tone colored faces

Finally, here is another rambling exploration of shadows when black/white images are replaced by texture, including a voiceover. The point is that visual system can link disconnected parts of a surface when they are separated by a shadow. For example, if a shadow falls across a bright region, it creates two bright areas separated by a dark one. Normally, this is no problem and vision connects the bright region and dismisses the dark swath as an accident of illumination. However, when the same image is rendered as equiluminous (grey and dotted textures), this linking fails, breaking apart disconnected regions.

Donald Duck, texture version