Abstract
Much of our understanding of motion perception derives from studies employing 2AFC judgments about the motion of simple grating or dot patterns. The validity of such an approach is predicated on subjects' errors being Gaussian-distributed and (largely) unbiased. However it is increasingly clear that the operation of the human visual system is optimized to guide real behaviour in real visual environments. To examine the validity of these assumptions using more natural stimuli and tasks, we had subjects make a judgment of the absolute direction of a translating fractal or natural scene (by adjusting the orientation of a subsequently viewed line). We analysed pooled direction-estimates to estimate the bias, variance, skew and kurtosis of error distributions as a function of motion direction. Our results broadly confirm several known phenomena such as the oblique effect for motion (OEM; best discrimination around cardinal directions), and reference repulsion (the tendency not to report motion in the cardinal directions). However we also report several new phenomena. First estimates are much more biased around oblique directions, but not greatly more variable, and this can contribute substantially to the OEM. Second, subjects' errors are not only somewhat more tightly distributed around the cardinals (the OEM) but are more leptokurtic (“peaky”/heavy-tailed), a property that violates the assumptions of psychophysical threshold estimation, elicits lower thresholds, and is likely to be linked to the orientation structure of natural scenes.
Supported by Sydney University and the Wellcome Trust.