When you inspect the above image you may notice that the central diamond moves, shifts or “swims” within / above the background. I call this the Orthogonal Dotted Lines Sway.
∆ Angle) of the diamond lines vs. background lines and is (of course) lost when the lines are parallel
Shakeangle influences illusion strength. More on this below.
Equilum.toggles “equiluminance” (setting the dashed black-white to red-green of same brightness); equiluminance abolishes the seeming motion
Shadowcheckbox demonstrates that a seeming depth does not interact with the motion (prompted by the shadow some of Kitaoka’s examples)
Orthogonal Dotted Lines Swayallows to test other patterns.
Colbutton is just for aesthetic reasons (i.e. fun)
You are seeing your own eye movement here. The phenomenon belongs to the class of relative motion illusions, including the Ōuchi, the “Spine drift” illusion and examples by Akiyoshi Kitaoka, namely “Kite” and “Kite 2”. With the pop-up that defaults to
Orthogonal Dotted Lines Sway, these and various other patterns can be selected. They all evoke shifting, swimming or swaying relative motion illusions, to different degrees; the
Orthogonal Dotted Lines Sway is possibly the simplest arrangement to evoke it, yet quite strong.
The “integration bias hypothesis” goes back to Hine (1997) but is best explained (and supported by data) by Mather (2000). Basically, if an rectangle is moved obliquely, its short and long edges generate motion signals, which have to be vectorially combined to recover the correct direction. Any bias leads to misestimation of the correct direction. Here motion direction along the dotted line sees high edge contrast (evoking a strong motion signal), while orthogonal to it the contrast is lower (grey background), causing a weaker motion signal. When the full pattern is moved (e.g. by eye movements), the misestimation of the central disk is orthogonal to the one of the surround, leading to relative displacement, thus to the illusion. As a test: When the background is darkened (use the lower of the 3 “color wells”), leading to more equal motion signals from all edges, the illusion is gone.
I built the present figure based on this hypothesis to explain this rather beautiful phenomenon (unfortunately, this hypothesis does not explain the “maximal sway direction”, so it fails):
While nothing seems actually wrong with the above :), the motion blur hypothesis predicts the wrong sway/shift angle. In the case of “Orthogonal Dotted Lines Sway”, along or 90° orthogonal to the lines is predicted, but (easy to test) a 45° shift (relative to lines, here corresponds to 0° or 90°) produces the strongest illusion. Bummer.