Primitives used in the spatial localization of nonabutting stimuli: peaks or centroids

In order to determine whether simple luminance profiles are located by their peaks or centroids we performed a three element alignment task where the central element's degree of luminance asymmetry was randomly chosen from a flat distribution (skew noise). The central element with its randomly chosen skew was either positioned using the peak or centroid of its distribution. Accuracy is invariant with the magnitude of the skew noise for the centroid but not the peak condition. We conclude that the human visual system assigns position tags using centroids not peaks of luminance distributions for gabors. However this is not the case for Gaussian blobs, where a measure closer to the midpoint is used for our stimulus arrangement.