Motion-surface labeling by orientation, spatial frequency and luminance polarity in 3-D structure-from-motion

A compelling percept of three-dimensionality is attainable from a purely motion-defined simulation of a transparent rotating cylinder, referred to as 3-D structure-from-motion (SFM). Interestingly, subjects rarely perceive reversals of the cylinder's direction of rotation when they are introduced. Treue, Andersen, Ando, and Hildreth (Vision Res. 35 (1995) 139-148) have argued that this reflects the visual system's insensitivity to the textural detail on the cylinder's motion surfaces. We have recently shown however that with cylinders made from oriented micropatterns, motion reversals are perceived when the orientations of the micropatterns are different on the cylinder's front/back surfaces, suggesting that the visual system is sensitive to the type of feature in these stimuli (Vision Res. 39 (1999) 881-886). In the present study we extended this finding by testing for feature-sensitivity along other dimensions besides orientation, specifically spatial frequency, colour and luminance polarity. We found that subjects perceived more rotation direction reversals when the front/back surfaces of the cylinder were segregated, as opposed to non-segregated by feature-type, along all of these dimensions except, notably, colour. We also investigated the stage at which the feature-sensitivity is incorporated in 3-D SFM. We reasoned that if 3-D SFM mechanisms were tuned, or labeled for feature-type, swapping of features during the cylinder's rotation would result in illusory reversals in just the feature-segregated condition, whereas if grouping of like-features preceded the formation of 3-D motion surfaces, no such illusory reversals would be expected. We found that feature-swapping resulted in more illusory reversals in the feature-segregated compared to non-segregated conditions, supporting the mechanism tuning, or labeling, hypothesis.