Abstract
Process that the visual system produces summary statistical representations from any feature is called ensemble perception. Average orientation perception has been often studied by two-dimensional display of visual stimuli with homogeneous properties. However, in a three-dimensional environment, properties often vary on the retina even if the stimuli are identical. Binocular disparity occurs on the retina during depth perception and the near stimuli have larger and lower spatial frequency retinal images than the same stimuli at the far location. It is unclear how much the variability of each non-target feature (i.e., binocular disparity, size, and spatial frequency of retinal images) influences average estimation of a target feature (i.e., orientation). In this study, we manipulated the variability of each non-target feature independently under a stereoscopic environment to investigate their interference on average orientation estimation. We conducted an experiment with the following four display conditions: (1) eight Gabor patches of 0.94 cycles/degree (cpd) or 0.5 cpd at the standard plane (i.e., a physical display), (2) four patches at the standard plane and another four patches at a far plane (the binocular disparity was 5.8 arcmin), (3) 4-high and 4-low frequency patches at the standard plane (0.94 cpd and 0.5 cpd), (4) 4-large and 4-small patches at the standard plane (about 2.2 degrees and 1.2 degrees of visual angle). Other retinal properties were always equal. The task was to indicate average direction of the eight patches as right or left to the vertical. Variability level of the eight orientations differed in every trial. After calculation of average orientation threshold of each display condition, the statistically significant threshold elevation was observed only in the size variability condition, suggesting that variability of retinal size might have loaded on orientation integration more than the variability of binocular disparity and spatial frequency.