Abstract
Estimation of any visual property typically involves pooling information over an extended image region, but such pooling is useful only if the region selected is appropriate. Successful estimation thus depends critically on perceptual segmentation. Previous work has demonstrated that the perceived orientation of dot clusters is predicted by their principal axis. We investigated the interaction of perceptual segmentation and the perceived orientation of dot clusters that could potentially be segmented into two sub-clusters.
Methods. Stimuli were dot clusters formed by the union of a large sub-cluster (uniformly sampled within an ellipse) and a small circular sub-cluster. Variables manipulated were: distance of the small sub-cluster from the main axis of the ellipse; dot density; and radius of the small sub-cluster. Dot clusters were presented for 1sec and masked. Observers adjusted the orientation of a pattern consisting of multiple parallel lines, to match the perceived orientation of the dot cluster.
Results. As the separation of the sub-clusters increased, perceived orientation shifted gradually from the global principal axis (of both sub-clusters merged) to that of the larger sub-cluster alone. Thus with increased separation—hence increased likelihood of segmentation—the dots within the smaller sub-cluster are given systematically lower weights in the principal axis computation. Moreover, the shift occurred sooner for higher dot densities—consistent with the expectation that these require smaller separations to be reliably segmented. The visual system thus employs a strategy of robust statistics, wherein data points are weighted differentially based on the probability that they arose from a separate generative process.