Purchase this article with an account.
David Kane, Peter Bex, Steven Dakin; Humans assume isotropic orientation structure when solving the ‘aperture problem’ for motion. Journal of Vision 2010;10(7):832. doi: https://doi.org/10.1167/10.7.832.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
We examined how global direction judgements with stimuli prone to the “aperture problem” depend on the local orientation structure of the stimulus. Observers adjusted the orientation of a line to match the overall direction of four randomly positioned Gabors whose carrier velocities were consistent with rigid-translation in a single random direction. The four Gabor orientations were either randomly distributed or evenly spaced at 45° intervals. Response variability was ∼20° in the evenly spaced condition and ∼30° in the random orientation condition. The degree of correlation between observers' errors when retested with identical stimuli was greater in the random orientation condition, demonstrating that the increase in variability is almost entirely determined by trial-by-trial differences in the orientation structure of the stimuli. In contrast the majority of the errors (∼80%) in the evenly spaced condition are random. Because two or more different velocities uniquely specify a particular global direction, an ideal-observer that fits a cosine to the local velocity distribution will not produce errors, while adding random noise will produce unpredictable errors (unlike human observers). However when the motion energy model is incorporated as a local motion stage, the representation of each local velocity is no longer discrete and the energy from differently oriented elements may overlap. Predictable errors may then arise from a mismatch between the local motion energy distribution (on a trial-by-trial basis) and a global motion stage that assumes an isotropic orientation structure (i.e. a cosine). The model now generates errors in the random orientation condition that correlate with the observers' errors (R2 ∼0.48). This compares to an R2 of ∼0.56 for the correlation between observers' test-retest errors, demonstrating that the model captures around 85% of the stimulus-driven variability.
This PDF is available to Subscribers Only