Free fuse the two columns of gray squares in (a), so that the left eye sees column L and the right eye sees column R (or vice versa: it doesnt matter), to give a single vertical column of gray squares. Note the perceived brightness of the squares. On a light background (a), the middle square (arrowed) is the same to both eyes and probably looks lighter than the squares above and below. However, on a black background (b) the middle square probably looks darker than those above and below. Squares in (a) and (b) are actually identical, only the backgrounds differ.
What s going on? The stimulus presents a light square to one eye and a dark square to the other eye, so that the two luminances always sum to a constant. So if your eyes (and your web browser) were linear, all fused squares would look the same brightness. However, the visual system is non-linear, and systematically overweights the square with the higher contrast (not luminance), favouring dark squares on a light background in (a), and favouring light squares on a dark background in (b). Careful measurements have shown that the weighting function is quadratic for light squares (spatial increments) but square, or winner take all, for dark squares (spatial decrements), as shown in the graphs.