Abstract
Previous findings indicate that observers estimate the center of dot clusters by averaging dot locations. More recent work with clusters made up of a large cluster and a partially-overlapping small cluster demonstrated that the visual system gives less weight to points in the small cluster to the extent that it can be segmented away from the large cluster (Cohen, Singh, Maloney, VSS2006; Juni, Singh, OPAM2006). Here we test whether the visual system is robust when dot clusters cannot be readily segmented. If it is, then points away from the center will receive less weight in estimating the center. Methods. Stimuli were comprised of a single isotropic bivariate Gaussian cluster containing 60 dots (SD=1.3°). The cluster appeared for 250ms at one of seven equally spaced locations horizontally. There were three conditions. In each we added 10 dots uniformly distributed in a small circular region (radius=1.09°) that was either 1.95° to the left, 1.95° to the right, or centered. Using a 2AFC task, observers indicated whether the entire cluster's center was to the right or left of a reference. Analysis. We fit psychometric functions to estimate PSEs for the three conditions. We used these to estimate the influence of adding dots at left, right and center—i.e., to estimate the weight given to the added dots. Four observers completed the experiment (3×7×50=1050 trials). Results. Two observers did not exhibit any significant robustness: they assigned equal weight to points near the center and away from the center of the distribution, effectively computing the average. The remaining two observers exhibited significant robustness with the average weight given the off-center added dots 0.38 and 0.66 of the weight assigned to the rest of the dots. Conclusion. At least some human observers use robust estimators in estimating the center of dot clusters with no evident part structure.
NSF BCS-0216944, NIH EY08266.