Evidence for Positional Coding in Hyperacuity

Observers performed simple pattern discriminations [tests of orientation (vernier) acuity and spatial-interval acuity] with targets consisting of spatially separated squares. We investigated the effects on acuity of supernumerary squares placed at various mean positions between the two squares constituting the target configuration. The exact position of the supernumerary squares relative to the target squares changed randomly from trial to trial, so that their spatial relationship to the targets could not serve as a cue. Observers attempted to ignore these supernumerary squares and to base their judgments on the outer target squares alone. The supernumerary squares raised thresholds if they were sufficiently close (4.4-arcmin separation) to the target squares but not if they were at a greater distance (21.0 arcmin). The results therefore show that the observers could ignore the supernumerary squares, even when they fell into the space between the target squares. This finding suggests that observers can select a class of length- and orientation-tuned filter that is suited exactly to the requirements of a particular psychophysical task. We argue that filter models of hyperacuity are insufficient unless they address this critical issue of filter selection and that a complete model requires an explicit spatial representation of target feature position.